Reinforcement Learning for Expert Reasoning: Founding Engineer

About Atman Labs
At Atman Labs we are building software to emulate human expertise. We believe our research poses a credible path to emulate true human cognition and interaction with deep knowledge and proactive reasoning, which has largely been impossible to do via standalone Artificial Intelligence techniques. Our unique research is inspired by biological priors, lies at the intersection of custom Reinforcement Learning environments and Large-Scale Knowledge Representation, and is evolved and compounded with commercial application. As an applied research and commercialization company, we are deploying our platform in products across a number of commercial domains to demonstrate the value of our approach – starting with building proactive shopping concierges for e-commerce, to eventually launching expert systems across travel, healthcare, education, science and more.

The Next Frontier of Reinforcement Learning: Emulating How Humans can Reason about Knowledge to Solve Complex, Long-Horizon Tasks
We are hiring for a founding engineer that will join the team working on building reinforcement learning environments to reason on structured knowledge representations, to mimic the cognitive abilities human experts have when solving complex tasks. As a specialist in reinforcement learning rooted in first-principled, biological thinking, you are able to discern that biological systems have infinite sequences of actions and rewards, which are also influenced by the knowledge and memories that these systems possess.

Today, reinforcement learning has only been deployed to solve tasks that are easily abstractable and don’t require understanding contextual knowledge, such as game-playing or navigation. However, for highly cognitive tasks that are stochastic in nature, require leveraging semantic knowledge, and have nearly infinite possibilities of actions – essentially every human expert task – the industry has yet to realize the power of reinforcement learning to model how humans can set goals and form strategies to achieve them by reasoning from their existing knowledge. The secret lies in the intersection of reinforcement learning and structured knowledge representations.

As a founding engineer focusing on expert planning, you will design custom reinforcement learning environments that (a) turn expert tasks into concrete and hierarchical goals, such as making shopping recommendations or teaching a challenging concept, (b) explore web-scale knowledge graphs to model the reasoning abilities of experts, in particular, how they use their existing knowledge to form strategies to solve those goals, and (c) interact with a variety of novel human-computer interfaces where feedback influences the system’s reward. Moreover, training reinforcement learning agents in human-centric applications will require large amounts of human interaction data. Generating infinite data by building interaction simulation models is an exciting avenue you will help tackle, intersecting with generative models and infrastructure design.

About You
We are looking for ambitious and independent thinkers who have a deep desire to contribute and want to be part of the team that makes this a reality for humanity. Our founding team prides itself in building a highly conscious culture that intentionally uplifts each member towards their personal ambitions while fostering collective innovation. To contribute to and grow alongside our team, you should either have the following qualities, or be willing to rise to them with our support:

Technical

• You have a PhD or equivalent industrial expertise in the application of reinforcement learning with knowledge of the limitations and frontiers of the field.
• You have demonstrated expertise in designing, testing, and deploying end-to-end reinforcement learning solutions, with continuous environments, and can help set the guidelines for both experimentation and production deployment.
• You are deeply proficient in various policy- or value-based RL methods, from Proximal Policy Optimization (PPO), Deep Q Networks (DQN), and/or Monte Carlo methods, and can strategically experiment with and select between various methods to fit the current situation or intended behavior.
• You have some intuition and interest in knowledge graphs, GNNs, and graph embeddings, and are eager to explore how continuous representations of knowledge can be connected to a continuous state space, action space, and reward that best mimics how experts solve cognitive tasks.
• You have 7+ years of programming experience in Python and have development experience with both ML toolkits and RL environments like OpenAI Gymnasium. You are equally capable as a software engineer as you are in formulating novel research ideas and your code proves it.

Personal

• You are capable of reasoning from first-principles, where there is no trodden path, as well as critically evaluate when existing ideas are worth considering.
• You are articulate and can present your ideas in writing, in person, and in small groups, and are able to educate audiences at all levels on the novel applications and relevance of reinforcement learning.
• You are eager to amplify a team that has deep research and product experience, coupled with a bold vision for the future of intelligent systems.
• You can easily distinguish authentic and high integrity thinkers from ‘posers’, while also critically evaluating truth from fiction in your own work.
• Your colleagues consider you a highly positive personality, you amplify the energy of others rather than dampen the mood.
• Your intensity goes from 0 to 1000 when you become authentically interested in a topic.
• You not only have interests in reinforcement learning, but are deeply curious about a range of interdisciplinary topics, ranging from knowledge graphs, recommendations, web-scale search, deep learning, generative AI models, computer vision and the opportunity to build truly intelligent systems in software that are inspired by biology.
• You can show high creativity and intensity in your personal pursuits, and your intelligence, creativity, and motivation is not limited to only one discipline.
• You consider yourself an innovator and an original thinker, not a follower. You are looking for a way to contribute to the world, and want to join our team to do so.
• You want to work in person in London. Don’t worry, we’ll sponsor your visa.

We have the ambition to usher the world towards co-existing alongside Benevolent AGI.
Not only do we believe that our work is a credible approach to functionally emulate the intelligence of human experts in machines, but we believe that this mission can also allow us to conceive many commercial products that yield billions of dollars of commercial revenues to support an ambitious R&D effort for years to come. We are building for a future where humans coexist alongside benevolent AGI and we will be at the forefront of executing on this vision. We are looking for ambitious and independent thinkers who have a deep desire to contribute and want to be part of the team that makes this a reality for humanity.

We’re excited to meet you. If you are too, send a short message with a list of your projects and highlights, as well as a brief paragraph of your life’s story, to [email protected].