Research Scientist, Machine Learning, Scalable Alignment

At DeepMind, we value diversity of experience, knowledge, backgrounds and perspectives, and harness these qualities to create extraordinary impact. We are committed to equal employment opportunity regardless of sex, race, religion or belief, ethnic or national origin, disability, age, citizenship, marital, domestic or civil partnership status, sexual orientation, gender identity, pregnancy, maternity or related condition (including breastfeeding) or any other basis as protected by applicable law. If you have a disability or additional need that requires accommodation, please do not hesitate to let us know.

Snapshot

At DeepMind, we've built a unique culture and work environment where long-term ambitious research can flourish. Our special interdisciplinary team combines the best techniques from deep learning, reinforcement learning and systems neuroscience to build general-purpose learning algorithms. We have already made a number of high profile breakthroughs towards building artificial general intelligence, and we have all the ingredients in place to make further significant progress over the coming year!

About us

We’re a dedicated scientific community, committed to “solving intelligence” and ensuring our technology is used for widespread public benefit.

We have an encouraging and inclusive environment where collaboration is encouraged and learning is shared freely.

We constantly iterate on our workplace experience to ensure it encourages a balanced life. Our list of benefits is extensive, and we’re happy to discuss this further throughout the interview process!

The team

The goal of the Scalable Alignment Team (SAT) is to make highly capable agents do what humans want, even when it is difficult for humans to know what that is. This means we want to remove subtle biases, factual errors, or deceptive behaviour even if they would normally go unnoticed by humans, whether due to reasoning failures or biases in humans or due to very capable behaviour by the agents.

To achieve this, we ask humans what they want and train agents to do that, assisting humans in judgements by providing evidence, outlining arguments, and pointing out subtleties. As language is a key medium for human communication, much of SAT’s work revolves around large language models (LLMs) such as Chinchilla, fine-tuning these models using techniques such as human preference RL, debate, citing evidence, and LM red teaming. We view LLMs both as a tool for safety by enabling human-machine communication and as examples of ML models that may cause both near-term and long-term harms. Since our goal is to do what humans want, the uncertainties involved are about humans, not just ML; we need to carefully design the interaction between humans and machines to achieve answers humans would endorse after careful reflection.

We view human interaction as only one component of safety, and work with many other teams at DeepMind to build a unified overall strategy, including Alignment, Ethics and Society, and Strategy and Governance.

The role

Scalable Alignment Research Scientists at DeepMind seek to design, implement, and understand training algorithms for communication agents that learn from human interaction, with the dual goals of using communication as a tool for safety and of mitigating harms caused by these models.

Having pioneered research in the world's leading academic and industrial labs in PhDs, post-docs or professorships, Research Scientists join DeepMind to work collaboratively within and across research fields. They develop solutions to fundamental questions, drawing on expertise from a variety of fields.

Key Responsibilities

• Improve our understanding of what it means for an agent to do what humans want even in subtle cases, and design research plans and evaluation schemes that accurately measure this.
• Design, implement, and understand training processes for agents that learn from human interaction, incorporating ideas from machine learning, cognitive science, and related areas.
• Collaborate with research teams externally and internally to ensure that SAT’s work acts as a piece of a unified safety plan for AGI.
• Report and present research findings and developments to internal and external collaborators with effective written and verbal communication.

About you

We look for the following skills and experience:

• PhD in a technical field, or equivalent practical experience.

In addition, the following would be an advantage:

• PhD in machine learning, computer science, statistics, computational neuroscience, mathematics, or physics.
• Research experience in and/or technical knowledge of AI alignment or natural language processing.
• A passion for making AGI go well.

Competitive salary applies.