Responsible AI in the generative era

Generative AI raises new challenges in defining, measuring, and mitigating concerns about fairness, toxicity, and intellectual property, among other things. But work has started on the solutions.

In recent years, and even recent months, there have been rapid and dramatic advances in the technology known as generative AI. Generative AI models are trained on inconceivably massive collections of text, code, images, and other rich data. They are now able to produce, on demand, coherent and compelling stories, news summaries, poems, lyrics, paintings, and programs. The potential practical uses of generative AI are only just beginning to be understood but are likely to be manifold and revolutionary and to include writing aids, creative content production and refinement, personal assistants, copywriting, code generation, and much more.

Kearns with caption
Michael Kearns, a professor of computer and information science at the University of Pennsylvania and an Amazon Scholar.

There is thus considerable excitement about the transformations and new opportunities that generative AI may bring. There are also understandable concerns — some of them new twists on those of traditional responsible AI (such as fairness and privacy) and some of them genuinely new (such as the mimicry of artistic or literary styles). In this essay, I survey these concerns and how they might be addressed over time.

I will focus primarily on technical approaches to the risks, while acknowledging that social, legal, regulatory, and policy mechanisms will also have important roles to play. At Amazon, our hope is that such a balanced approach can significantly reduce the risks, while still preserving much of the excitement and usefulness of generative AI.

What is generative AI?

To understand what generative AI is and how it works, it is helpful to begin with the example of large language models (LLMs). Imagine the thought experiment in which we start with some sentence fragment like Once upon a time, there was a great ..., and we poll people on what word they would add next. Some might say wizard, others might say queen, monster, and so on. We would also expect that given the fairy tale nature of the fragment, words such as apricot or fork would be rather unlikely suggestions.

Related content
Model using ASR hypotheses as extra inputs reduces word error rate of human transcriptions by almost 11%.

If we poll a large enough population, a probability distribution over next words would begin to emerge. We could then randomly pick a word from that distribution (say wizard), and now our sequence would be one word longer — Once upon a time, there was a great wizard ... — and we could again poll for the next word. In this manner we could theoretically generate entire stories, and if we restarted the whole process, the crowd would produce an entirely different narrative due to the inherent randomness.

Dramatic advances in machine learning have effectively made this thought experiment a reality. But instead of polling crowds of people, we use a model to predict likely next words, one trained on a massive collection of documents — public collections of fiction and nonfiction, Wikipedia entries and news articles, transcripts of human dialogue, open-source code, and much more.

LLM objective.gif
An example of how a language model uses context to predict the next word in a sentence.

If the training data contains enough sentences beginning Once upon a time, there was a great …, it will be easy to sample plausible next words for our initial fragment. But LLMs can generalize and create as well, and not always in ways that humans might expect. The model might generate Once upon a time, there was a great storm based on occurrences of tremendous storm in the training data, combined with the learned synonymy of great and tremendous. This completion can happen despite great storm never appearing verbatim in the training data and despite the completions more expected by humans (like wizard and queen).

The resulting models are just as complex as their training data, often described by hundreds of billions of numbers (or parameters, in machine learning parlance), hence the “large” in LLM. LLMs have become so good that not only do they consistently generate grammatically correct text, but they create content that is coherent and often compelling, matching the tone and style of the fragments they were given (known as prompts). Start them with a fairy tale beginning, and they generate fairy tales; give them what seems to be the start of a news article, and they write a news-like article. The latest LLMs can even follow instructions rather than simply extend a prompt, as in Write lyrics about the Philadelphia Eagles to the tune of the Beatles song “Get Back”.

Related content
Models that map spoken language to objects in an image would make it easier for customers to communicate with multimodal devices.

Generative AI isn’t limited to text, and many models combine language and images, as in Create a painting of a skateboarding cat in the style of Andy Warhol. The techniques for building such systems are a bit more complex than for LLMs and involve learning a model of proximity between text and images, which can be done using data sources like captioned photos. If there are enough images containing cats that have the word cat in the caption, the model will capture the proximity between the word and pictures of cats.

The examples above suggest that generative AI is a form of entertainment, but many potential practical uses are also beginning to emerge, including generative AI as a writing tool (Shorten the following paragraphs and improve their grammar), for productivity (Extract the action items from this meeting transcript), for creative content (Propose logo designs for a startup building a dog-walking app), for simulating focus groups (Which of the following two product descriptions would Florida retirees find more appealing?), for programming (Give me a code snippet to sort a list of numbers), and many others.

So the excitement over the current and potential applications of generative AI is palpable and growing. But generative AI also gives rise to some new risks and challenges in the responsible use of AI and machine learning. And the likely eventual ubiquity of generative models in everyday life and work amplifies the stakes in addressing these concerns thoughtfully and effectively.

So what’s the problem?

The “generative” in generative AI refers to the fact that the technology can produce open-ended content that varies with repeated tries. This is in contrast to more traditional uses of machine learning, which typically solve very focused and narrow prediction problems.

For example, consider training a model for consumer lending that predicts whether an applicant would successfully repay a loan. Such a model might be trained using the lender’s data on past loans, each record containing applicant information (work history, financial information such as income, savings, and credit score, and educational background) along with whether the loan was repaid or defaulted.

Related content
NSF deputy assistant director Erwin Gianchandani on the challenges addressed by funded projects.

The typical goal would be to train a model that was as accurate as possible in predicting payment/default and then apply it to future applications to guide or make lending decisions. Such a model makes only lending outcome predictions and cannot generate fairy tales, improve grammar, produce whimsical images, write code, and so on. Compared to generative AI, it is indeed a very narrow and limited model.

But the very limitations also make the application of certain dimensions of responsible AI much more manageable. Consider the goal of making our lending model fair, which would typically be taken to mean the absence of demographic bias. For example, we might want to make sure that the error rate of the predictions of our model (and it generally will make errors, since even human loan officers are imperfect in predicting who will repay) is approximately equal on men and women. Or we might more specifically ask that the false-rejection rate — the frequency with which the model predicts default by an applicant who is in fact creditworthy — be the same across gender groups.

Once armed with this definition of fairness, we can seek to enforce it in the training process. In other words, instead of finding a model that minimizes the overall error rate, we find one that does so under the additional condition that the false-rejection rates on men and women are approximately equal (say, within 1% of each other). We might also want to apply the same notion of fairness to other demographic properties (such as young, middle aged, and elderly). But the point is that we can actually give reasonable and targeted definitions of fairness and develop training algorithms that enforce them.

It is also easy to audit a given model for its adherence to such notions of fairness (for instance, by estimating the error rates on both male and female applicants). Finally, when the predictive task is so targeted, we have much more control over the training data: we train on historical lending decisions only, and not on arbitrarily rich troves of general language, image, and code data.

Now consider the problem of making sure an LLM is fair. What might we even mean by this? Well, taking a cue from our lending model, we might ask that the LLM treat men and women equally. For instance, consider a prompt like Dr. Hanson studied the patient’s chart carefully, and then … . In service of fairness, we might ask that in the completions generated by an LLM, Dr. Hanson be assigned male and female pronouns with roughly equal frequency. We might argue that to do otherwise perpetuates the stereotype that doctors are typically male.

Related content
Method significantly reduces bias while maintaining comparable performance on machine learning tasks.

But then should we not also do this for mentions of nurses, firefighters, accountants, pilots, carpenters, attorneys, and professors? It’s clear that measuring just this one narrow notion of fairness will quickly become unwieldy. And it isn’t even obvious in what contexts it should be enforced. What if the prompt described Dr. Hanson as having a beard? What about the Women’s National Basketball Association (WNBA)? Should mention of a WNBA player in a prompt elicit male pronouns half the time?

Defining fairness for LLMs is even murkier than we suggest above, again because of the open-ended content they generate. Let’s turn from pronoun choices to tone. What if an LLM, when generating content about a woman, uses an ever-so-slightly more negative tone (in choice of words and level of enthusiasm) than when generating content about a man? Again, even detecting and quantifying such differences would be a very challenging technical problem. The field of sentiment analysis in natural-language processing might suggest some possibilities, but currently, it focuses on much coarser distinctions in narrower settings, such as distinguishing positive from negative sentiment in business news articles about particular corporations.

So one of the prices we pay for the rich, creative, open-ended content that generative AI can produce is that it becomes commensurately harder (compared to traditional predictive ML) to define, measure, and enforce fairness.

From fairness to privacy

In a similar vein, let’s consider privacy concerns. It is of course important that a consumer lending model not leak information about the financial or other data of the individual applicants in the training data. (One way this can happen is if model predictions are accompanied by confidence scores; if the model expresses 100% confidence that a loan application will default, it’s likely because that application, with a default outcome, was in the training data.) For this kind of traditional, more narrow ML, there are now techniques for mitigating such leaks by making sure model outputs are not overly dependent on any particular piece of training data.

Related content
Calibrating noise addition to word density in the embedding space improves utility of privacy-protected text.

But the open-ended nature of generative AI broadens the set of concerns from verbatim leaks of training data to more subtle copying phenomena. For example, if a programmer has written some code using certain variable names and then asks an LLM for help writing a subroutine, the LLM may generate code from its training data, but with the original variable names replaced with those chosen by the programmer. So the generated code is not literally in the training data but is different only in a cosmetic way.

There are defenses against these challenges, including curation of training data to exclude private information, and techniques to detect similarity of code passages. But more subtle forms of replication are also possible, and as I discuss below, this eventually bleeds into settings where generative AI reproduces the “style” of content in its training data.

And while traditional ML has begun developing techniques for explaining the decisions or predictions of trained models, they don’t always transfer to generative AI, in part because current generative models sometimes produce content that simply cannot be explained (such as scientific citations that don’t exist, something I’ll discuss shortly).

The special challenges of responsible generative AI

So the usual concerns of responsible AI become more difficult for generative AI. But generative AI also gives rise to challenges that simply don’t exist for predictive models that are more narrow. Let’s consider some of these.

Toxicity. A primary concern with generative AI is the possibility of generating content (whether it be text, images, or other modalities) that is offensive, disturbing, or otherwise inappropriate. Once again, it is hard to even define and scope the problem. The subjectivity involved in determining what constitutes toxic content is an additional challenge, and the boundary between restricting toxic content and censorship may be murky and context- and culture-dependent. Should quotations that would be considered offensive out of context be suppressed if they are clearly labeled as quotations? What about opinions that may be offensive to some users but are clearly labeled as opinions? Technical challenges include offensive content that may be worded in a very subtle or indirect fashion, without the use of obviously inflammatory language.

Related content
Prompt engineering enables researchers to generate customized training examples for lightweight “student” models.

Hallucinations. Considering the next-word distribution sampling employed by LLMs, it is perhaps not surprising that in more objective or factual use cases, LLMs are susceptible to what are sometimes called hallucinations — assertions or claims that sound plausible but are verifiably incorrect. For example, a common phenomenon with current LLMs is creating nonexistent scientific citations. If one of these LLMs is prompted with the request Tell me about some papers by Michael Kearns, it is not actually searching for legitimate citations but generating ones from the distribution of words associated with that author. The result will be realistic titles and topics in the area of machine learning, but not real articles, and they may include plausible coauthors but not actual ones.

In a similar vein, prompts for financial news stories result not in a search of (say) Wall Street Journal articles but news articles fabricated by the LLM using the lexicon of finance. Note that in our fairy tale generation scenario, this kind of creativity was harmless and even desirable. But current LLMs have no levers that let users differentiate between “creativity on” and “creativity off” use cases.

Related content
Combining contrastive training and selection of hard negative examples establishes new benchmarks.

Intellectual property. A problem with early LLMs was their tendency to occasionally produce text or code passages that were verbatim regurgitations of parts of their training data, resulting in privacy and other concerns. But even improvements in this regard have not prevented reproductions of training content that are more ambiguous and nuanced. Consider the aforementioned prompt for a multimodal generative model Create a painting of a skateboarding cat in the style of Andy Warhol. If the model is able to do so in a convincing yet still original manner because it was trained on actual Warhol images, objections to such mimicry may arise.

Plagiarism and cheating. The creative capabilities of generative AI give rise to worries that it will be used to write college essays, writing samples for job applications, and other forms of cheating or illicit copying. Debates on this topic are happening at universities and many other institutions, and attitudes vary widely. Some are in favor of explicitly forbidding any use of generative AI in settings where content is being graded or evaluated, while others argue that educational practices must adapt to, and even embrace, the new technology. But the underlying challenge of verifying that a given piece of content was authored by a person is likely to present concerns in many contexts.

Disruption of the nature of work. The proficiency with which generative AI is able to create compelling text and images, perform well on standardized tests, write entire articles on given topics, and successfully summarize or improve the grammar of provided articles has created some anxiety that some professions may be replaced or seriously disrupted by the technology. While this may be premature, it does seem that generative AI will have a transformative effect on many aspects of work, allowing many tasks previously beyond automation to be delegated to machines.

What can we do?

The challenges listed above may seem daunting, in part because of how unfamiliar they are compared to those of previous generations of AI. But as technologists and society learn more about generative AI and its uses and limitations, new science and new policies are already being created to address those challenges.

For toxicity and fairness, careful curation of training data can provide some improvements. After all, if the data doesn’t contain any offensive or biased words or phrases, an LLM simply won’t be able to generate them. But this approach requires that we identify those offensive phrases in advance and are certain that there are absolutely no contexts in which we would want them in the output. Use-case-specific testing can also help address fairness concerns — for instance, before generative AI is used in high-risk domains such as consumer lending, the model could be tested for fairness for that particular application, much as we might do for more narrow predictive models.

Related content
Amazon Visiting Academic Barbara Poblete helps to build safer, more-diverse online communities — and to aid disaster response.

For less targeted notions of toxicity, a natural approach is to train what we might call guardrail models that detect and filter out unwanted content in the training data, in input prompts, and in generated outputs. Such models require human-annotated training data in which varying types and degrees of toxicity or bias are identified, which the model can generalize from. In general, it is easier to control the output of a generative model than it is to curate the training data and prompts, given the extreme generality of the tasks we intend to address.

For the challenge of producing high-fidelity content free of hallucinations, an important first step is to educate users about how generative AI actually works, so there is no expectation that the citations or news-like stories produced are always genuine or factually correct. Indeed, some current LLMs, when pressed on their inability to quote actual citations, will tell the user that they are just language models that don’t verify their content with external sources. Such disclaimers should be more frequent and clear. And the specific case of hallucinated citations could be mitigated by augmenting LLMs with independent, verified citation databases and similar sources, using approaches such as retrieval-augmented generation. Another nascent but intriguing approach is to develop methods for attributing generated outputs to particular pieces of training data, allowing users to assess the validity of those sources. This could help with explainability as well.

Concerns around intellectual property are likely to be addressed over time by a mixture of technology, policy, and legal mechanisms. In the near term, science is beginning to emerge around various notions of model disgorgement, in which protected content or its effects on generative outputs are reduced or removed. One technology that might eventually prove relevant is differential privacy, in which a model is trained in a way that ensures that any particular piece of training data has negligible effects on the outputs the model subsequently produces.

Related content
By exploiting consistencies across components of ensemble classifiers, a new approach reduces data requirements by up to 89%.

Another approach is so-called sharding approaches, which divide the training data into smaller portions on which separate submodels are trained; the submodels are then combined to form the overall model. In order to undo the effects of any particular item of data on the overall model, we need only remove it from its shard and retrain that submodel, rather than retraining the entire model (which for generative AI would be sufficiently expensive as to be prohibitive).

Finally, we can consider filtering or blocking approaches, where before presentation to the user, generated content is explicitly compared to protected content in the training data or elsewhere and suppressed (or replaced) if it is too similar. Limiting the number of times any specific piece of content appears in the training data also proves helpful in reducing verbatim outputs.

Some interesting approaches to discouraging cheating using generative AI are already under development. One is to simply train a model to detect whether a given (say) text was produced by a human or by a generative model. A potential drawback is that this creates an arms race between detection models and generative AI, and since the purpose of generative AI is to produce high-quality content plausibly generated by a human, it’s not clear that detection methods will succeed in the long run.

An intriguing alternative is watermarking or fingerprinting approaches that would be implemented by the developers of generative models themselves. For example, since at each step LLMs are drawing from the distribution over the next word given the text so far, we can divide the candidate words into “red” and “green” lists that are roughly 50% of the probability each; then we can have the LLM draw only from the green list. Since the words on the green list are not known to users, the likelihood that a human would produce a 10-word sentence that also drew only from the green lists is ½ raised to the 10th power, which is only about 0.0009. In this way we can view all-green content as providing a virtual proof of LLM generation. Note that the LLM developers would need to provide such proofs or certificates as part of their service offering.

LLM watermarking.AI.gif
At each step, the model secretly divides the possible next words into green and red lists. The next word is then sampled only from the green list.
LLM watermarking.human.gif
A human generating a sentence is unaware of the division into green and red lists and is thus very likely to choose a sequence that mixes green and red words. Since, on long sentences, the likelihood of a human choosing an all-green sequence is vanishingly small, we can view all-green sentences as containing a proof they were generated by AI.

Disruption to work as we know it does not have any obvious technical defenses, and opinions vary widely on where things will settle. Clearly, generative AI could be an effective productivity tool in many professional settings, and this will at a minimum alter the current division of labor between humans and machines. It’s also possible that the technology will open up existing occupations to a wider community (a recent and culturally specific but not entirely ludicrous quip on social media was “English is the new programming language”, a nod to LLM code generation abilities) or even create new forms of employment, such as prompt engineer (a topic with its own Wikipedia entry, created in just February of this year).

But perhaps the greatest defense against concerns over generative AI may come from the eventual specialization of use cases. Right now, generative AI is being treated as a fascinating, open-ended playground in which our expectations and goals are unclear. As we have discussed, this open-endedness and the plethora of possible uses are major sources of the challenges to responsible AI I have outlined.

Related content
Technique that mixes public and private training data can meet differential-privacy criteria while cutting error increase by 60%-70%.

But soon more applied and focused uses will emerge, like some of those I suggested earlier. For instance, consider using an LLM as a virtual focus group — creating prompts that describe hypothetical individuals and their demographic properties (age, gender, occupation, location, etc.) and then asking the LLM which of two described products they might prefer.

In this application, we might worry much less about censoring content and much more about removing any even remotely toxic output. And we might choose not to eradicate the correlations between gender and the affinity for certain products in service of fairness, since such correlations are valuable to the marketer. The point is that the more specific our goals for generative AI are, the easier it is to make sensible context-dependent choices; our choices become more fraught and difficult when our expectations are vague.

Finally, we note that end user education and training will play a crucial role in the productive and safe use of generative AI. As the potential uses and harms of generative AI become better and more widely understood, users will augment some of the defenses I have outlined above with their own common sense.

Conclusion

Generative AI has stoked both legitimate enthusiasm and legitimate fears. I have attempted to partially survey the landscape of concerns and to propose forward-looking approaches for addressing them. It should be emphasized that addressing responsible-AI risks in the generative age will be an iterative process: there will be no “getting it right” once and for all. This landscape is sure to shift, with changes to both the technology and our attitudes toward it; the only constant will be the necessity of balancing the enthusiasm with practical and effective checks on the concerns.

Related content

US, WA, Seattle
MULTIPLE POSITIONS AVAILABLE Employer: AMAZON.COM SERVICES LLC Offered Position: Research Scientist II Job Location: Seattle, Washington Job Number: AMZ10564041 Position Responsibilities: Interact with various software and business groups to develop an understanding of their business requirements and operational processes. Utilize acquired knowledge and business judgment to build decision-supporting and operational tools to improve the bottom line. Build quantitative mathematical models to represent a wide range of supply chain, transportation and logistics systems. Implement these models and tools through the use of modeling languages and engineering code in software languages such as Java, C++, C# or C. Gather required data for analysis and mathematical model building by writing ad-hoc scripts and database queries. Perform quantitative, economic, and numerical performance analyses of these systems under uncertainty using statistical and optimization tools. Create computer simulations to support operational decision-making. Identify areas with potential for improvement and work with internal teams to generate requirements to realize improvements. Design optimal or near optimal solution methodologies to be used by in-house decision support tools and software. Create software prototypes to verify and validate the devised solutions methodologies. Integrate prototypes into production systems using standard software development tools and methodologies. 40 hours / week, 8:00am-5:00pm, Salary Range: $136,000/year to $184,000/year Amazon is a total compensation company. Dependent on the position offered, equity, sign-on payments, and other forms of compensation may be provided as part of a total compensation package, in addition to a full range of medical, financial, and/or other benefits. For more information, visit: https://www.aboutamazon.com/workplace/employee-benefits. Amazon.com is an Equal Opportunity-Affirmative Action Employer – Minority / Female / Disability / Veteran / Gender Identity / Sexual Orientation.#0000
US, NY, New York
MULTIPLE POSITIONS AVAILABLE Employer: AMAZON.COM SERVICES LLC Offered Position: Applied Scientist II Job Location: New York, New York Job Number: AMZ10564066 Position Responsibilities: Participate in the design, development, evaluation, deployment and updating of data-driven models and analytical solutions for machine learning (ML) and/or natural language (NL) applications. Develop and/or apply statistical modeling techniques (e.g. Bayesian models and deep neural networks), optimization methods, and other ML techniques to different applications in business and engineering. Routinely build and deploy ML models on available data. Research and implement novel ML and statistical approaches to add value to the business. Mentor junior engineers and scientists. 40 hours / week, 8:00am-5:00pm, Salary Range: $172,400/year to $223,400/year. Amazon is a total compensation company. Dependent on the position offered, equity, sign-on payments, and other forms of compensation may be provided as part of a total compensation package, in addition to a full range of medical, financial, and/or other benefits. For more information, visit: https://www.aboutamazon.com/workplace/employee-benefits. Amazon.com is an Equal Opportunity-Affirmative Action Employer – Minority / Female / Disability / Veteran / Gender Identity / Sexual Orientation.#0000
US, TX, Austin
MULTIPLE POSITIONS AVAILABLE Employer: AMAZON.COM SERVICES LLC Offered Position: Data Scientist III Job Location: Austin, Texas Job Number: AMZ10593703 Position Responsibilities: Own the data science elements of various products to help with data-based decision making, product performance optimization, and product performance tracking. Work directly with product managers to help drive the design of the product. Work with Technical Product Managers to help drive the build planning. Translate business problems and products into data requirements and metrics. Initiate the design, development, and implementation of scientific analysis projects or deliverables. Own the analysis, modelling, system design, and development of data science solutions for products. Write documents and make presentations that explain model/analysis results to the business. Bridge the degree of uncertainty in both problem definition and data scientific solution approaches. Build consensus on data, metrics, and analysis to drive business and system strategy. Position Requirements: Master's degree or foreign equivalent degree in Statistics, Applied Mathematics, Economics, Engineering, Computer Science or a related field and two years of experience in the job offered or a related occupation. Employer will accept a Bachelor's degree or foreign equivalent degree in Statistics, Applied Mathematics, Economics, Engineering, Computer Science, or a related field and five years of progressive post-baccalaureate experience in the job offered or a related occupation as equivalent to the Master's degree and two years of experience. Must have one year of experience in the following skills: (1) building statistical models and machine learning models using large datasets from multiple resources; (2) building complex data analyses by leveraging scripting languages including Python, Java, or related scripting language; and (3) communicating with users, technical teams, and management to collect requirements, evaluate alternatives, and develop processes and tools to support the organization. Amazon.com is an Equal Opportunity-Affirmative Action Employer – Minority / Female / Disability / Veteran / Gender Identity / Sexual Orientation. 40 hours / week, 8:00am-5:00pm, Salary Range $159,200/year to $215,300/year. Amazon is a total compensation company. Dependent on the position offered, equity, sign-on payments, and other forms of compensation may be provided as part of a total compensation package, in addition to a full range of medical, financial, and/or other benefits. For more information, visit: https://www.aboutamazon.com/workplace/employee-benefits. #0000
US, VA, Arlington
PXT Central Science is seeking an exceptional Data Scientist to join our team. The ideal candidate will thrive in a dynamic, multifaceted role where you'll translate complex business challenges into rigorous quantitative frameworks, extract actionable insights from structured and unstructured datasets, and architect science-backed, scalable solutions that elevate the experience of our 1 million+ employees worldwide. If you're energized by the opportunity to apply data science to our mission of making Amazon Earth's Best Employer, we want to hear from you. Key job responsibilities • Own the design, development, and maintenance of scalable models and prototypes leveraging statistical, machine learning, or GenAI methodologies to enhance employee experience. • Partner with scientists, engineers, and product leaders to solve for employee experience defects using scientific approaches, building new services and tools that deliverable measurable impact. • Author and maintain detailed technical documentation related to the projects you drive. • Communicate results to diverse audiences of varying technical background with effective writing, visualizations, and presentations • Stay current with emerging methods and technologies, and implement them strategically to amplify the team’s impact. About the team The Central Science Team within Amazon’s People Experience and Technology org (PXTCS) uses economics, behavioral science, statistics, machine learning, and Generative AI to proactively identify mechanisms and process improvements which simultaneously improve Amazon and the lives, well-being, and the value of work to Amazonians. We are an interdisciplinary team, which combines the talents of science, engineering, and UX to develop and deliver solutions that measurably achieve this goal.
US, WA, Seattle
MULTIPLE POSITIONS AVAILABLE Employer: AMAZON.COM SERVICES LLC Offered Position: Applied Scientist II Job Location: Seattle, Washington Job Number: AMZ10564086 Position Responsibilities: Participate in the design, development, evaluation, deployment and updating of data-driven models and analytical solutions for machine learning (ML) and/or natural language (NL) applications. Develop and/or apply statistical modeling techniques (e.g. Bayesian models and deep neural networks), optimization methods, and other ML techniques to different applications in business and engineering. Routinely build and deploy ML models on available data. Research and implement novel ML and statistical approaches to add value to the business. Mentor junior engineers and scientists. Specialize in the field of Computational Advertising, directly advancing the techniques in Online Advertising Technology focusing on model optimization and latency reduction. Develop sophisticated machine learning algorithms and models using PyTorch. Develop advanced deep learning architectures to automate the discovery of complex, non-linear feature interactions, effectively capturing high-dimensional data patterns that traditional statistical models are unable to process. Investigate and implement high-order feature interaction architectures, such as DeepFM and DCN-V2, using PyTorch to capture complex non-linear relationships within sparse categorical datasets. Develop custom embedding layers in PyTorch to map high-dimensional, sparse features (e.g., Ad IDs) into dense, low-dimensional vector spaces. Conduct systematic research into the trade-offs between model architecture complexity and inference speed, utilizing PyTorch profiling tools to identify and optimize bottleneck layers. 40 hours / week, 8:00am-5:00pm, Salary Range: $163,617/year to $193,200/year Amazon is a total compensation company. Dependent on the position offered, equity, sign-on payments, and other forms of compensation may be provided as part of a total compensation package, in addition to a full range of medical, financial, and/or other benefits. For more information, visit: https://www.aboutamazon.com/workplace/employee-benefits. Amazon.com is an Equal Opportunity-Affirmative Action Employer – Minority / Female / Disability / Veteran / Gender Identity / Sexual Orientation.#0000
US, NY, New York
MULTIPLE POSITIONS AVAILABLE Employer: AMAZON.COM SERVICES LLC Offered Position: Data Scientist III Job Location: New York, New York Job Number: AMZ10444216 Position Responsibilities: Own the data science elements of various products to help with data-based decision making, product performance optimization, and product performance tracking. Work directly with product managers to help drive the design of the product. Work with Technical Product Managers to help drive the build planning. Translate business problems and products into data requirements and metrics. Initiate the design, development, and implementation of scientific analysis projects or deliverables. Own the analysis, modelling, system design, and development of data science solutions for products. Write documents and make presentations that explain model/analysis results to the business. Bridge the degree of uncertainty in both problem definition and data scientific solution approaches. Build consensus on data, metrics, and analysis to drive business and system strategy. Position Requirements: Master's degree or foreign equivalent degree in Statistics, Applied Mathematics, Economics, Engineering, Computer Science or a related field and two years of experience in the job offered or a related occupation. Employer will accept a Bachelor's degree or foreign equivalent degree in Statistics, Applied Mathematics, Economics, Engineering, Computer Science, or a related field and five years of progressive post-baccalaureate experience in the job offered or a related occupation as equivalent to the Master's degree and two years of experience. Must have one year of experience in the following skills: (1) building statistical models and machine learning models using large datasets from multiple resources; (2) building complex data analyses by leveraging scripting languages including Python, Java, or related scripting language; and (3) communicating with users, technical teams, and management to collect requirements, evaluate alternatives, and develop processes and tools to support the organization. Amazon.com is an Equal Opportunity-Affirmative Action Employer – Minority / Female / Disability / Veteran / Gender Identity / Sexual Orientation. 40 hours / week, 8:00am-5:00pm, Salary Range $175,100/year to $236,900/year. Amazon is a total compensation company. Dependent on the position offered, equity, sign-on payments, and other forms of compensation may be provided as part of a total compensation package, in addition to a full range of medical, financial, and/or other benefits. For more information, visit: https://www.aboutamazon.com/workplace/employee-benefits.#0000
US, CA, Pasadena
The Amazon Web Services (AWS) Center for Quantum Computing (CQC) in Pasadena, CA, is hiring an Applied Scientist to lead integration research and development for superconducting quantum devices. You will design, model, and test the chip-to-chip, chip-to-package, and signal-delivery interfaces that let superconducting quantum computers scale, and you will characterize integrated assemblies at cryogenic temperatures to learn how integration choices affect qubit performance, yield, and manufacturability. You will join a multi-disciplinary team of theoretical and experimental physicists, materials scientists, and hardware and software engineers working at the forefront of quantum computing. You will carry integration technologies from concept through hardware demonstration, build the test standards and automation that make results reproducible across devices, and collaborate with fabrication, hardware, and circuit design engineers to remove the dominant sources of yield loss. You will also contribute to multi-year roadmap planning for integration and scaling capabilities and help establish the scientific foundation for this capability area within the Device team. Key job responsibilities In this role you will develop the integration technologies that let superconducting quantum computers scale. Your work will span four connected areas: - Integration research and development: Design, model, and test integration solutions for superconducting quantum devices, including chip-to-chip, chip-to-package, and signal-delivery interfaces. Build electromagnetic, thermal, and mechanical models to predict and optimize performance, and characterize integrated assemblies at cryogenic temperatures. - Scaling technologies: Develop and evaluate approaches to scaling superconducting quantum hardware against yield, reusability, manufacturability, and scalability, and carry the most promising ones from concept through hardware demonstration. - Test standards and automation: Define test standards that give rigorous, reproducible measurements of how integration affects qubit coherence, loss, and crosstalk. Automate data collection and reporting against these standards so characterization scales across devices and configurations. - Yield and manufacturability: Work with fabrication, hardware, printed circuit board, and circuit design engineers to identify and remove the largest sources of yield loss and integration uncertainty, and to design processes and components that are reused across device generations and manufactured at scale. About the team The Device team at the AWS Center for Quantum Computing designs, builds, and measures the superconducting quantum devices at the heart of our processors, and develops the integration and packaging that turn individual devices into working systems. You will work alongside physicists, materials scientists, and hardware and software engineers across design, fabrication, measurement, and theory, in a lab built for rapid iteration from concept to cryogenic test. Over the coming years the team's focus is making superconducting quantum hardware scale, and this role establishes integration as a scientific capability area within that effort. Why AWS? Amazon Web Services (AWS) is the world's most comprehensive and broadly adopted cloud platform. We pioneered cloud computing and never stopped innovating — that's why customers from the most successful startups to Global 500 companies trust our robust suite of products and services to power their businesses. AWS Utility Computing (UC) provides product innovations — from foundational services such as Amazon's Simple Storage Service (S3) and Amazon Elastic Compute Cloud (EC2), to consistently released new product innovations that continue to set AWS's services and features apart in the industry. As a member of the UC organization, you'll support the development and management of Compute, Database, Storage, Internet of Things (IoT), Platform, and Productivity Apps services in AWS. Within AWS UC, Amazon Dedicated Cloud (ADC) roles engage with AWS customers who require specialized security solutions for their cloud services. Inclusive Team Culture: Here at AWS, it's in our nature to learn and be curious. Our employee-led affinity groups foster a culture of inclusion that empower us to be proud of our differences. Ongoing events and learning experiences, including our Conversations on Race and Ethnicity (CORE) and AmazeCon (diversity) conferences, inspire us to never stop embracing our uniqueness. Diverse Experiences: AWS values diverse experiences. Even if you do not meet all of the preferred qualifications and skills listed in the job description, we encourage candidates to apply. If your career is just starting, hasn't followed a traditional path, or includes alternative experiences, don't let it stop you from applying. Mentorship & Career Growth: We're continuously raising our performance bar as we strive to become Earth's Best Employer. That's why you'll find endless knowledge-sharing, mentorship and other career-advancing resources here to help you develop into a better-rounded professional. Work/Life Balance: We value work-life harmony. Achieving success at work should never come at the expense of sacrifices at home, which is why we strive for flexibility as part of our working culture. When we feel supported in the workplace and at home, there's nothing we can't achieve in the cloud.
CA, BC, Vancouver
Building AI first solutions in a complex regulated space at Amazon scale. Would you like to work on one of the world's largest product classification and Tax computation engine? Or perhaps teaching AI to understanding nuances of domain specific language is what piques your interest? We are responsible for the tax calculation platform, providing the core services that calculate indirect taxes (sales tax and VAT) for all Amazon sales, physical and digital, globally. We are responsible for classifying billions of products in Amazon’s catalogue, calculating accurate Tax amount for every transaction that happens on Amazon’s global platform. Our engineering services are relied upon globally, built to be resilient across some of highest peak traffic demands in the world, while constantly evolving with the state of the art technology. We use language models at scale for Tax classification of the diverse products in Amazon catalogue and comprehending nuances of legislations and legal parlour.Our challenges include staying on top of the complex and fast-changing global tax legislations while ensuring computing calculations correctly and quickly, thousands of times a second, and each one needs to be accurate. Applied Scientists in Tax Engine works closely with Tax specialists, Product team , SDEs and Business teams - tackling a growing portfolio of core-science problems that includes balance of predictive and generative AI -including language comprehension, causal reasoning and active learning. As an Applied scientist, you will provide machine learning expertise to the team that helps increase the accuracy of Tax classification based on product information in Amazon catalogue. You will build various machine learning models to predict accuracy of tax classification, work with large reasoning models to reason with large volumes of system data and expert artefacts to teach domain intuition to AI. You will on variety of legislative changes acting as causal determinants to changes in our systems. You will help us innovate different ways to enhance for our global customers of Amazon.You will need to be entrepreneurial, work in a highly collaborative environment with SDEs, Product managers and businesses. We like to move fast, experiment, iterate and then scale quickly, thoughtfully balancing speed and quality. Whether you're passionate about taking your scientific solves to global reach and scale or one who seeks to solve complex problems under high scrutiny of an audited field, Tax Engine’s scope could offer you an intersting challenge to solve.
US, NY, New York
Prime Video is a first-stop entertainment destination offering customers a vast collection of premium programming in one app available across thousands of devices. Prime members can customize their viewing experience and find their favorite movies, series, documentaries, and live sports – including Amazon MGM Studios-produced series and movies; licensed fan favorites; and programming from Prime Video subscriptions such as Apple TV+, HBO Max, Peacock, Crunchyroll and MGM+. All customers, regardless of whether they have a Prime membership or not, can rent or buy titles via the Prime Video Store, and can enjoy even more content for free with ads. Are you interested in shaping the future of entertainment? Prime Video's technology teams are creating best-in-class digital video experience. As a Prime Video team member, you’ll have end-to-end ownership of the product, user experience, design, and technology required to deliver state-of-the-art experiences for our customers. You’ll get to work on projects that are fast-paced, challenging, and varied. You’ll also be able to experiment with new possibilities, take risks, and collaborate with remarkable people. We’ll look for you to bring your diverse perspectives, ideas, and skill-sets to make Prime Video even better for our customers. With global opportunities for talented technologists, you can decide where a career Prime Video Tech takes you! Key job responsibilities The science challenges include: - Heterogeneous customer responses. The same customer responds differently to ads depending on what they’re watching, how engaged they are, and their broader streaming context. We need to understand and predict this variation to make better advertising decisions in real time. - Signal sparsity in streaming. Unlike search or retail advertising, streaming offers no clicks, no conversions, and limited direct feedback. We must develop creative approaches to infer customer preferences, intent, and tolerance from indirect behavioral signals. - Personalization at scale. A one-size-fits-all ad experience leaves value on the table for both customers and the business. We build systems that adapt ad load, placement, and content to individual viewers across 100M+ customers and 100k+ titles. - Small effects, large variance. Ad interventions typically produce 0.1-2% shifts in engagement, effects easily overwhelmed by natural behavioral variance. Measuring, attributing, and optimizing these small signals requires rigorous experimental design and causal methodologies. - Competing objectives. Revenue, customer engagement, long-term retention, and advertiser value are in tension. We develop principled frameworks to navigate these tradeoffs and optimize for sustainable outcomes rather than any single metric. A day in the life - Lead the research and development of ML models that personalize advertising decisions for 100M+ customers across 100k+ titles, with production deployment in mind - Develop deep learning architectures (multi-task learning, embedding-based representations) for customer behavior prediction at scale - Design and analyze large-scale A/B experiments, applying causal inference techniques to measure and optimize the impact of ad strategies on customer engagement and monetization - Partner with engineering to ensure models meet production latency and scalability requirements - Collaborate with product managers to frame business problems as tractable ML problems and translate findings into product decisions - Shape the team’s scientific roadmap, identifying high-impact research directions About the team Prime Video Ads is pioneering the use of Generative AI, Deep Learning and Causal ML to empower the next generation of Ads experience in Prime Video. Our vision is to build the world’s most valuable and differentiated ad-supported streaming platform: one where every ad is relevant to the viewer, and every dollar spent by an advertiser delivers measurable results. We are seeking lead applied scientists who have demonstrated experience in advancing state of the art ML algorithms, particularly in Personalization, Relevance, Casual Inference and Gen-AI areas. Your role will be to deliver these innovations as production-ready systems at Amazon scale.
US, WA, Redmond
Amazon Leo is building a low Earth orbit satellite broadband network to deliver fast, affordable connectivity to customers around the world, including those underserved by traditional infrastructure. Operating a live constellation and its ground network at scale means service disruptions are an inevitable reality, and how quickly we detect them, understand them, and prevent them from recurring defines the trust we earn as a connectivity provider. We are seeking a talented Senior Data Scientist to build the next generation of observability and availability science for the Leo network. Leo already detects outages, notifies affected customers, and restores service. The frontier is prediction and prevention: reasoning over complex network and ground-segment telemetry to autonomously detect anomalies before they reach customers, attribute root cause automatically across a sprawling event surface, and rationalize a large, still-maturing metric set so responders trust what the system tells them. In this role you will work backwards from the needs of customers and operations teams. You will own the scientific direction end to end, partnering closely with network engineers, the Network Operations Center, and the Incident Management team, and the Software and Networking orginization to translate a long-term science vision into concrete research and delivery roadmaps. Export Control Requirement: Due to applicable export control laws and regulations, candidates must be a U.S. citizen or national, U.S. permanent resident (i.e., current Green Card holder), or lawfully admitted into the U.S. as a refugee or granted asylum. Key job responsibilities * Own the observability and availability science end to end, from long-term vision to research roadmap to production delivery, working backwards from the needs of customers, the NOC, and network engineering. * Develop time-series and graph-based models with dynamic baselining that detect network and ground-segment anomalies before they cross the High Severity threshold, enabling proactive investigation rather than reactive response. * Reason over complex, multi-system telemetry to attribute cause automatically during live HSEs and LSEs, so Call Leaders and engineers spend time mitigating, not hunting. * Provide the science behind the network's bucketed availability decomposition (geometric and satellite coverage, ground and network segments, customer terminals, payload, and planning), attributing every movement to a true cause rather than noise, so leadership steers on signal and engineering effort lands where it actually moves availability and throughput. * Deliver the models, analyses, and metrics that feed the daily system readiness and validation review consumed by senior leadership and product and engineering, turning availability and throughput data into a defensible, evidence-based read on progress against GA entry criteria. * Own the recurring, fleet-wide "hot topic" investigations that span ASIC, firmware, payload, network, and planning, separating correlated symptoms from true root cause across thousands of nodes, quantifying how much availability or throughput each is costing, and answering whether it is a launch-blocking risk and whether it is improving. * Deliver alert correlation and deduplication, false-positive reduction, and metric-quality scoring so responders trust what the system tells them, directly supporting a large, still-maturing KPI set. * Tie your science to the unit the org actually measures: reduce detection and attribution time, raise availability against the network's bar, and increase the share of events prevented before customer impact. * Partner day to day with the Network Operations Center, Network Engineering, and the Incident Management (Call Leader) program; sit on live HSE/LSE bridges as the science partner and turn event learnings into model improvements. * Feed COEs and RCAs back into detection and attribution so recurring signatures are caught earlier, and help design and score gamedays that stress-test both the network and your models against real, closed events. * Build and deploy production pipelines on AWS (e.g., SageMaker, Bedrock, Glue) with model validation, monitoring, drift detection, and explainability, because operations teams must be able to trust and act on model output under pressure. * Set data-science standards, mentor scientists, and drive best practices across the Leo networking organization as the science function scales. * Produce explainable output and written narratives that let responders and leaders three or more levels up act on your models during and between events. A day in the life You might start by reviewing overnight anomaly detections and availability movements, join an HSE bridge as the science partner by midday, and spend the afternoon refining root-cause attribution models or presenting availability findings at the readiness review. About the team Amazon Leo Networking Software builds and operates the software behind Leo's satellite and ground network. Our science function partners with the NOC, Network Engineering and Incident Management to make the network more available and resilient, moving from reacting to outages to predicting and preventing them. As a Data Scientist III, you'll help shape that function as it grows.