VERSES AI Research
@VERSESAI
Followers
1,834
Following
58
Media
0
Statuses
128
VERSES AI research team advancing cognitive computing and next-generation AI inspired by nature @helloVERSES
Joined November 2022
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
#DNC2024
• 168441 Tweets
River
• 145298 Tweets
Macri
• 89559 Tweets
Bill Clinton
• 84986 Tweets
Simón
• 79175 Tweets
#虎に翼
• 56039 Tweets
Okan
• 44881 Tweets
Mike Lindell
• 39555 Tweets
Talleres
• 25798 Tweets
サンサン
• 25228 Tweets
JENNA
• 24677 Tweets
#AEWDynamite
• 22521 Tweets
Soto
• 22116 Tweets
Oprah
• 21028 Tweets
Borja
• 18482 Tweets
Peñarol
• 16877 Tweets
Stevie Wonder
• 14902 Tweets
Geoff Duncan
• 14229 Tweets
Joey Votto
• 12205 Tweets
Pinned Tweet
🚀🚨🤖 Introducing the VERSES Research AI white paper 🚀🚨🤖
Short thread on "Designing Ecosystems of Intelligence from First Principles" 🧵 1/13
5
46
144
🚨🤖 Preprint on explainability and AI from
@VERSESAI
🧠🚨 “Designing explainable artificial intelligence with active inference,” by
@MahaultAlbarra1
,
@ineshipolito
, Safae Essafi Tremblay,
@JasonGFox
,
@GReal1111
, Karl Friston, and
@mjdramstead
1/7
1
8
33
New from
@VERSESAI
and collaborators: introducing "inductive planning," a new form of active inference allowing us to generate intelligent agents that aim to achieve explicit goals
New formulation of active inference just dropped: meet inductive planning
1
9
50
2
6
32
More exciting news to end the week: Volvo joins NASA and other key players in the Genius™ Beta Program
News! "Volvo Joins VERSES Genius™ Beta Program"
Collaboration Centered on Assisted and Autonomous Driving Safety
Read the press releases:
#autonomousdriving
#agi
#genius
#verses
1
12
36
1
5
24
🚀🚨🤖 New candidate search 🚀🚨🤖 We’re seeking an AI Ethicist at the VERSES
@helloVERSES
Research Lab
@VERSESAI
Short thread 🧵
Apply here:
1
6
18
We highlight the crucial role of shared narratives, goals, and communication in the emergence of collective behavior—and we discuss how active inference helps account for this in terms of sharing (aspects of) the same generative model 10/13
1
0
17
The result is AI that "scales up" the way nature does: by aggregating individual intelligences and their locally contextualized knowledge bases into "nested intelligences" (rather than by simply adding parameters or layers to a model) 11/13
1
0
17
We close with a sketch of stages of development for AI using active inference, culminating in "shared" intelligences that emerge from a network of cognitively and emotionally intelligent (or "sympathetic") agents able to take one another's perspectives 13/13
0
1
16
We consider ensuing ethical issues, considering the limits of technical solutions. We argue that while not a panacea, active inference helps to make certain aspects of AI more auditable—and that successful shared intelligence must reflect our most essential and core values 12/13
1
0
15
We argue that active inference is uniquely suited to the engineering of ecosystems of intelligence, because it allows one to write down the dynamics of sparsely coupled systems that self-organize over several scales 9/13
1
0
15
Our core thesis is that the design of intelligent systems must begin from the physicality of information and its processing at every scale or level of self-organization, and that taking the multi-scale aspects of intelligence seriously will be transformative for AI research 4/13
1
0
15
Active inference furnishes a mathematical approach to intelligence, understood as generating evidence for your own existence (aka self-evidencing); and to collective intelligence, formalized as shared (self-) evidencing premised on shared narratives and goals 7/13
1
0
14
The self-evidencing paradigm motivates our using (Bayesian) model evidence (or its lower bound) as an objective function, allowing us to model homeostasis and allostasis as preference-driven drives, and promoting curiosity in complex agents 8/13
1
0
14
Accordingly, our vision for AI is based on active inference, which combines the virtues of a "first-principles," physics-based approach to AI with Bayesian formulations—thus reframing and, in some key respects, extending Bayesian state-of-the-art approaches in AI 5/13
1
0
14
Active inference can be read as a physics of intelligence, which inherits from the physics of self-organization. It entails an explicit mechanics of the beliefs of agents and groups of agents—known as Bayesian mechanics—which is inherently multi-scale 6/13
1
0
13
We propose a roadmap towards the development of a distributed ecosystem of intelligence in which both humans and human-designed intelligences are integral participants—what we call "shared intelligence" 3/13
1
0
13
We're proud to share with you the VERSES Research Lab artificial intelligence white paper, which presents our vision of research and development in the field of AI, premised on using active inference as a core technology 2/13
1
0
13
Exciting news!
[NEWS] VERSES Increases Previously Announced Overnight Marketed & Private Placement Offering to $20 Million and Announces Pricing of Units and Special Warrants
Visit our website for the full release 🔗
$VERS
#AI
#ArtificialIntelligence
#Financing
2
7
23
0
1
8
The future of AI depends on our ability to build trust and shared understanding. By embracing design principles from active inference, we aim to foster trust, decrease the risk of harm, and identify and mitigate biases and discriminatory patterns in AI decision-making 7/7
0
0
5
🔥 Hot off the presses 🔥
🚨 Our new flagship paper on the free energy principle & Bayesian mechanics has now been published! 🚨
“On Bayesian mechanics: a physics of and by beliefs” by
@DaltonSakthi
@conorheins
@KoudahlM
@BerenMillidge
@lancelotdacosta
@jkbren
and Karl Friston 1/7
6
67
207
0
4
5
This capacity, powered by active inference, to access and report on their own beliefs and belief-updating processes—when combined with explicit, interpretable modeling—facilitates a transparent, clear, and tractable understanding of AI decision-making 6/7
1
0
5
By now, it is a truism that AI is poised to have a transformative effect on several core domains of human activity. But state of the art AI systems are “black boxes”: their decision-making processes are difficult to interpret, explain, and audit by users and stakeholders 2/7
1
0
5
We argue that active inference can be a game-changer for developing explainable AI systems. More specifically, we propose that the shift to designing AI systems based on active inference opens the door to greatly enhances explainability of AI—and this, for two reasons 3/7
2
0
5
We’re looking for candidates with in-depth knowledge of ethical theories and principles—especially as related auditability and explainability, privacy, awareness of bias, fairness, and accountability—as well as in-depth knowledge of AI implementation and deployment 4/5
1
0
4
This is a full time, fully remote position. Global compensation packages are very competitive and staff enjoy generous PTO. For US based employees, VERSES provides robust company paid health programs. We’re a fun, exciting, and diverse workplace. Please share widely! 5/5
0
0
4
Second, active inference can inform the design of AI systems able to access, track, and report on their own inferences. Active inference has been used to model self-access and self-report with great success. We leverage these technologies to enhance AI (self-) explainability 5/7
1
0
3
VERSES is looking for an AI Ethics specialist to help lead our research in AI Ethics. In particular, the candidate will be writing and implementing ethical guidelines and standards for AI research, development, and deployment at scale at VERSES 2/5
1
0
4
First, the key to building explainable AI systems lies in the core architecture. We propose a novel architecture for AI systems that—because it is centered on an explicit generative mode—enables users and stakeholders to explain and audit the factors influencing its decisions 4/7
1
0
3
They will be expected to identify and manage risks associated with AI projects, to collaborate with legal and technical stakeholders, to ensure that ethical principles are being followed in our AI research projects, and will play a crucial role in the design of our AI systems 3/5
1
0
4