Strategy Robot, Inc.

Strategy Robot CEO Dr. Tuomas Sandholm will be speaking on the "AI for National Security: What’s the Urgency?" panel at the AI Horizons Pittsburgh Summit on Monday at 2-3 PM US Eastern time. https://lnkd.in/eJAVZzVS We will also have a table at the Summit and we look forward to talking to you there! We are hiring AI and software developers at the PhD, MS, and BS levels, with experience and straight out of school. US citizens only. You can contact us through our web page strategyrobot.ai.

Some had trouble downloading our press release about our world-leading fighter pilot AIs. Here's a shortened version: Strategy Robot has been awarded a contract to apply its Course-of-Action Generation and Execution System (CGES) for many-v-many beyond-visual-range air combat tactics under DARPA's AI Reinforcements (AIR) program. The Phase 1 contract totals $6M over 18 months. The AIR program takes AI-based aerial combat to a new level compared to prior demonstrations that have focused on 1-v-1 within-visual-range combat, further assuming perfect observation of the opponent. Such scenarios are called dogfights; they started in World War I. In contrast, modern air combat takes place outside of visual range, has multiple aircraft per side, and they can only observe enemies via sensors, which yield an imperfect view. Strategy Robot is in the AIR program’s Technical Area 2, which seeks to develop “dominant tactical autonomy for multi-ship, beyond visual range air combat missions.” Furthermore, “Autonomy solutions will initially be developed and demonstrated on human-on-the-loop F-16 testbeds and then transferred to an uncrewed combat aerial vehicle.” The autonomy part is less than 1% of the task: Strategy Robot’s AI learned to fly fighter aircraft in one day of self-training in a simulator. The hard 99% is optimizing the tactics against an enemy who is also optimizing its tactics. Computational game theory AI is the only sound way of planning against intelligent adversaries. It is the only methodology capable of generating non-exploitable strategies, by taking into account that one side’s best strategy depends on the other side’s strategy, and vice versa. CGES is unique in that it is the only game-theoretic Course of Action (COA) generation system, enabling optimal COA development under imperfect information. It has already proven superhuman during extensive testing in defense applications such as large-scale armored combat, portfolio planning, multi-base defense, and missile warfare. Our AIs are also multi-time world champions in international AI-v-AI benchmarks for imperfect-information games such as Texas hold’em, and have proven superhuman there too. “One could get some of the easy first steps done to an extent with heuristic hard-coded autonomy approaches and machine learning, but only game-theoretic techniques can get all the way to where this needs to go,” said Dr. Tuomas Sandholm, Founder, President, and CEO. “It is the only way an AI can soundly understand the adversary’s actions and inactions as signals of their private information. It is also the only way an AI can soundly understand how its own actions and inactions signal its private information to the adversary. Even top human pilots don’t know how to do it super well. It is extremely complicated. CGES significantly helps pilots generate and execute such tactics. Furthermore, the COAs that it generates are human understandable, unlike those from deep reinforcement learning.”

We are finally allowed to say a little bit about our world-leading fighter pilot AI: https://lnkd.in/eFSpa8H5

We are finally allowed to say a little bit about our world-leading fighter pilot AI: https://lnkd.in/eFSpa8H5

Exciting news! Strategy Robot is moving to a 6x bigger, state-of-the-art office tomorrow to support our growth. All bootstrapped!

Strategy Robot CEO will be conducting interviews at the NeurIPS-23 conference. AI & game theory for defense. Apply for positions at https://lnkd.in/gn35QMUx