Monday, January 24, 2022

Review of 2021 & Plans for 2022

SUMMARY:

Keen team - significant growth
Space Engineers - two major updates released
AI Game - looking promising
Memetic Badger

It is my tradition at the beginning of every year to publish a review of what we have achieved in the previous year, and what our plans are for the new one.

Review from last year (2020/2021) is here.

This year's review starts here. Enjoy!

Keen Software House

2021 was a year of changes but especially a year of growth, not only the team size, but also of our skills and experience in various projects.

In 2021, we proved that our studio is compatible with remote work and have continued to hire new team members from around the world.
A growing team demands changes in structure, and so many of the veteran team members were promoted to leadership positions.
Space Engineers: we released two major updates

Warfare 1: our New Warfare series launched with Warfare 1: Field Engineer and engineer combat saw massive changes and updates - read more

Heavy Industry: read more

Cooperation with our community: Space Engineers has more than 440,000 items on Steam workshop. We started to support the UGC (User generated content) sharing in cooperation with the Tebex team.
We added new official languages to Space Engineers: Italian, Brazilian - Portuguese, and Spanish.
We launched regular promotional events on our www.SpaceEngineersGame.com website, which are always supported by regular live game streams run by Joel.
Our programmers used Manual Badger architecture to evaluate parallelization techniques that would benefit the future development of Space Engineers. The ultimate aim of the investigation was to improve game performance to better leverage multi-core CPUs and GPUs. – article
We launched an updated affiliate program in cooperation with Xsolla.
We are dedicated for the continued development of Space Engineers on PC and Xbox
We collaborated with Makeship and over 1100 community members got their hands on an exclusive Space Engineer plushie! https://www.makeship.com/products/space-engineers-plush

We introduced crossplay and brought our Xbox and PC communities closer. https://blog.marekrosa.org/2021/02/space-engineers-community-crossplay.html
We’ve been supporting our community playing on the platform of their choice by pushing simultaneous releases and regular updates to both platforms PC & Xbox.
We are now continuing that series with Warfare 2: Broadside which focuses on massive changes to ship combat.
https://www.spaceengineersgame.com/new-players/warfare2-combat-guide/

GoodAI

GoodAI Research

The research team continued to focus on our multi-agent Badger architecture. We developed several implementations focused on learning through backpropagation. We investigated properties of inner/outer-loop meta-learning and converted three results into several papers that are now being submitted to scientific conferences.

The second half of the year was dedicated to memetic badger, a flavor of badger architecture inspired by cultural evolution theory and the ease with which science has progressed in the recent millenia (with no apparent change in the human genome). Memetic badger focuses on using and evolving memes (algorithms, data) for learning to solve problems. At present, the system is in active development and will remain the focus of research in 2022.

Papers and blog posts

Memetic Experiments for an Open-Ended Learning Paradigm - paper

Building a research culture where we share questions and difficulties, not just answers - article

GoodAI Grants & EU Projects

Our grants program began in October 2020 with the goal to support research that is relevant to the most pressing questions related to artificial intelligence on a whole and, more specifically, to GoodAI's Badger architecture.

Since the first award, we've had a great response from the research community around the world and granted nine more in 2021.

Over $777,000 has been awarded in total. Funding also supported the publication of papers on AI development:

We’re entering the second year of our work on EU grant projects:

The VeriDream project is an international consortium of six organizations across Europe implementing AI into robotics. Significant progress was made on our case study applying quality diversity algorithms to control robotic legs.

We are also part of iv4XR, a Horizon 2020 project that aims to build a novel verification and validation technology for Extended Reality systems based on techniques from AI to provide learning and reasoning over a virtual world.

Workshops & other news:

We held a 5-day seminar, Beyond Life-long Learning via Modular Meta-Learning, bringing together the GoodAI research team, grants recipients, and other experts to share knowledge on the pathway to a life-long learning system (GoodAI Research Roadmap 2021/2022) - article

GoodAI’s Senior Research Scientist, ‪Jaroslav Vítků authored a handy tool for the management of experiments in AWS (Amazon Web Services).

We participated in and hosted a hybrid event for the ALIFE 2021 conference - article

GoodAI Scientist Nicholas Guttenberg won the Evocraft Minecraft Open-Endedness Challenge.

Our workshop, From Cells to Societies, was accepted to the 2022 ICLR Conference.

AI Ethics & Safety

We continued to work towards the development of a safe AI for individuals and society.

Inceptional UNESCO agreement on an ethical framework for AI development - article
European Commission Report on Humans and Society in the Age of AI – article
AI in the Czech Republic – Marek Rosa panel discussion with the Aspen Institute Central Europe.
Cyborg Soldiers and Bioethics – Marek Rosa interview with GLOBSEC
AI & Happiness Roundtable – Marek Havrda summarizes thoughts from the meeting – article

AI Game

A new project aiming to find synergy between the AI and game development expertise present in GoodAI and Keen began.

As part of this project, we’ve been developing a new game (working title ‘AI Game’) that allows players to explore a world inhabited by intelligent NPCs powered by state-of-the-art AI.

The goal is to push video game boundaries in terms of immersion, emotions, dialogues and flexibility.

Preliminary work on the AI Game actually started towards the end of 2019. Since then we have:

Iterated through numerous prototypes
Developed a custom AI Engine
Built a well-functioning team of professionals (10 and growing)
Gained a lot of know-how and experience.
The game is starting to be entertaining, but more iteration are needed

Oranžérie

Oranžérie is the head-quarter of Keen and GoodAI.

In the Spring, we will begin a complete remodeling of the garden, adding pavilions, chill-out areas, a proper pavement, and a gate.

Pavilion in progress

Space Engineers Apotheosis - more details

COVID

Even as we collectively enter into the third year of the pandemic, COVID has not been an obstacle to company growth.

Our hiring reach expanded globally and our team is currently composed of 18 nationalities in Keen and 15 in GoodAI.

At present, 64% of our team is located in the Czech Republic and 36% is abroad.

For comparison, last year it was 75% in the Czech Republic and 25% abroad.

Number of newcomers in 2021:

GoodAI: 15
Keen: 25

Personal

The main part of my time is dedicated to the production of Space Engineers - working on designs, technology, art, and overall direction of the game.

The second big part of my time is aimed at merging AI and game development, materialized in AI Game. I believe that AI is the future of game development - not just fancy NPCs, but also as a tool to use natural language to define an interactive universe that works.

The third part is about the development of Memetic Badger. I have observed that the interest in this subgenre of AI (AI system as a collective of cells and societies) has been growing in the last 2 years.

On a personal level, I work very hard on how to balance all these activities.

Plans for 2022

Keen Software House

Space Engineers - New update - Warfare 2: Broadside - will be released soon
Few more major updates to Space Engineers are planned for this year
We have many surprises, especially on the technology side of Space Engineers & VRAGE. We will be talking about them soon.

GoodAI

ICLR workshop: From Cells to Societies
Memetic Badger - more work is needed, especially on integration between internal evolution inside the agent and the challenges of the external environment
AI Game - I am looking forward to some of the technology advancements we are planning in the next months. Also, we will be growing our ‘language model / transformers’ team.

Thank you for reading this blog!

Best,
Marek Rosa
CEO, Creative Director, Founder at Keen Software House
CEO, CTO, Founder at GoodAI

For more news:
Space Engineers: www.SpaceEngineersGame.com
Keen Software House: www.keenswh.com
Medieval Engineers: www.MedievalEngineers.com
GoodAI: www.GoodAI.com
General AI Challenge: www.General-AI-Challenge.org

Personal bio:
Marek Rosa is the CEO and CTO of GoodAI, a general artificial intelligence R&D company, and the CEO and founder of Keen Software House, an independent game development studio best known for its best-seller Space Engineers (4 million copies sold). Both companies are based in Prague, Czech Republic.

Marek has been interested in artificial intelligence since childhood. He started his career as a programmer but later transitioned to a leadership role. After the success of the Keen Software House titles, Marek was able to personally fund GoodAI, his new general AI research company building human-level artificial intelligence.

GoodAI was founded in January 2014, with a $10 Million investment from Marek, it now has over 30 research scientists, engineers, and consultants working across its divisions.

At this time, Marek is developing Space Engineers, as well as leading daily research and development on recursive self-improvement based general AI architecture - Badger.

Monday, January 17, 2022

What are the differences between centralized learning (in monolithic systems) and decentralized learning (in multi-agent systems)?

SUMMARY:

Question: Why are we studying social learning in multi-agent systems?
Answer: Multi-agent systems are made up of agents where each has its own objective. We believe that this leads to learning dynamics that are impossible in centralized systems.

This blog post is one of my personal takeaways from the Badger Seminar 2021.

Definitions:

Monolithic system (centralized): has one objective that is shared by all parts of the system. Let’s assume a homogeneous network and learning via a back-prop algorithm. This is what most deep learning is about.
Multi-agent system (decentralized): each part of the system has its own objective, therefore its learning resembles social learning in an evolving population of agents. We are not assuming back-prop learning.

Example of collective learning in a multi-agent system

An ant colony simulation where ants are stateless policies, moving randomly until they discover food, then start laying down pheromones (memories). If other ants discover pheromone trails, they start following them and adding more pheromones.

Summary: We can see that individual ants don’t know where the food is, but the collective of ants know where the food is and how to gather it.

Monolithic vs Multi-agent

The following table breaks down my intuitions. The hypotheses need to be verified experimentally.

Monolithic (centralized)	Multi-agent (decentralized, social learning)
Global objective. Centralized feedback.	Local objectives (don’t need to be aligned). Diverse objectives. Decentralized and local feedbacks.
	Could be a solution to “reward hacking” - because even if you hack your reward, you can’t hack the reward of other agents and their opinions on what your objectives should be
	Self-invented objectives possible. We want objectives (including the global one) to emerge, not to be predefined.
	Emergent building blocks.
	Progress within the system is induced by selection (competition), which is caused by limited resources (bottleneck, constraints).
Fixed credit mechanism.	Learned credit / feedback mechanism.
Serial learning.	Parallel learning.
Cost of communication doesn’t scale effectively in all-to-all networks.	Cost of communication doesn’t need to grow dramatically if we introduce modularization and hierarchical / heterarchical networking.
Every skill has to be learned individually, even if it’s needed in multiple places within the agent.	A learned skill can replicate within the society.
	This one isn’t clear and maybe I am completely wrong - but somehow I feel that the society of agents will consume less storage than one monolithic system, especially if they learn to reuse skills in modular fashion. The opposite is also quite possible 😉 and then the answer is “what if storage and parallel execution aren’t our bottlenecks, but time is?”
Fixed learning procedure.	Learned learning procedures.
Open-endedness is not possible because learning converges. Converging to one solution.	Open-endedness is possible because learning diverges. Interactions between parts of the system. Parts of the system are liberated from other parts. We can get diverse solutions, unlike in monolithic systems, where the training converges to just one solution.
Homogenous learning policy.	Heterogeneous learning policy.
Upper bound Learning of a monolithic system may be faster and/or leading to better task performance only up to a certain threshold (task complexity, number of agent’s parameters necessary for the task, etc). After this threshold, a transition to social learning will be necessary, otherwise no further task performance improvements are possible, or learning will take unreasonable time, or the adaptation to new tasks won’t be possible at all.	No upper bound.
	Dynamic topologies. Different topology for learning and inference passes.
New skills are learned only during the global backprop pass.	Horizontal and vertical transfer of learned skills. Replication is easier.
Fully connected systems are slower learners if there’s a cost for connections.
	Collective will generalize better to novel tasks than monolithic would. It can sacrifice parts to specialized tasks and keep the rest to stay better at adapting.

Conclusion

Emergent learning is a type of learning that can only happen on the collective level when multiple individuals are interacting and qualitatively new behaviors can emerge.

The main difference between monolithic and multi-agent systems: the latter is made from agents where each has local objectives (not a global objective) and this leads to interesting evolutionary dynamics.

Questions for reflection

Key points of social learning:

External information storage - Is it the key for better collective learning? The storage can be cumulative and bigger than the memory of an individual agent.
Multiple feedback mechanisms - A social system can have many adaptive feedback mechanisms, will they scale better than a centralized one in monolithic systems?
Efficiency threshold - Is there a threshold at which social systems become more efficient than monolithic systems?

Identified benefits of social systems:

Better scaling - does a modular / hierarchical system with mostly local communication scale better than a monolithic system?
Replication of skills - discovered skills can be replicated to other parts of the society, whereas in a monolithic system it needs to be rediscovered. Are there some counterexamples?
Open-ended learning - due to not having a single fixed feedback mechanism and the learning diverges, the social systems are more suitable for open-ended learning.

General:

Does a monolithic system learn faster than a multi-agent system?
Is there a limit where a monolithic system won’t be sufficient anymore and you need to switch to multi-agent learning? Can we get open-ended learning inside a monolithic system?
Can we simulate a multi-agent system on a monolithic system ? For example, a multi-agent system being simulated by a monolithic interpreter.

Thank you for reading this blog!

Best,
Marek Rosa
CEO, Creative Director, Founder at Keen Software House
CEO, CTO, Founder at GoodAI

GoodAI was founded in January 2014, with a $10 Million investment from Marek, it now has over 30 research scientists, engineers, and consultants working across its divisions.

At this time, Marek is developing Space Engineers, as well as leading daily research and development on recursive self-improvement based general AI architecture - Badger.