An interesting claim to AGI
A Canandian cognitive computing company ‘Verses AI’ specializing in biologically inspired distributed intelligence claims it has developed AGI. The company has invited Open AI to collaborate (and not compete) with it based on the Charter principles of Open AI issued in the pursuit of non-competitive AGI development. The invitation was issued through a full-page newspaper advertisement.
The company’s discovery seems to be a superior method to implement AGI due to the claims of tractability, lesser data and power consumption which the other AI computing models like Chat GPT, Bard and others clearly lack. However, these are ‘claims’ at the present moment pending verification.
“The VERSES AI Research Lab has spent years fine-tuning the intricacies surrounding Active Inference,” said Dr. Friston. “By incorporating the Free Energy Principle, a physics-based approach to AI, we are granted the possibility of modeling any physical system as an Active Inference agent that reasons, learns, plans and predicts in real-time.”
“Following the Free Energy Principle — which reconciles our brains’ predictions about the world with reality to reduce the differences between the two — Active Inference codifies mathematical principles that explicitly link intelligence, cognition and rational behavior to physical processes at any scale, from macroscopic to microscopic. With this framework, VERSES aims to solve probabilistic AI’s tractability problem, allowing users to design and deploy adaptive, real-time, scalable AI. ” Read the press release.
The AI Black box is not pitch black anymore
Finally, tiny cracks begin to appear admitting slivers of light into the AI ‘Black Box’. In a heartening development, awaited with bated breath, Anthropic’s researchers including Chris Olah who have been working to reverse engineer the way neural networks function, have discovered logical neural network behaviour which could satisfy parts of the ‘explainability’ criterion so essential but hitherto lacking for AI design and development. The technique applied is called ‘mechanistic interpretability’. The polysemantic behaviour of neurons, their sequence of their firing and their multi-modal capability are at the heart of the explanation. The three factors combined provide neural networks with the awesomely intelligent method to encode the humongous number of parameters of LLMs. Using Sparse Auto Encoders and Decoders to verify their findings, the Anthropic team has made a big contribution to further our understanding of AI. Read more here
AI News Update
The changing of the year brought on some expected legal challenges for some Big Tech companies. AI accountability is coming to the fore finally it seems.
AI Accountability Jan 5 (Reuters) – OpenAI and its financial backer Microsoft (MSFT.O) were sued on Friday in Manhattan federal court by a pair of nonfiction authors who say the companies misused their work to train the artificial-intelligence models behind the popular chatbot ChatGPT and other AI-based services. Writers Nicholas Basbanes and Nicholas Gage told the court in a proposed class action that the companies infringed their copyrights by including several of their books as part of the data used to train OpenAI’s GPT large language model. Read more here .
Indian Government Skills Certification In a POSITIVE step, the Indian Government plans to recognise Skill Certification of MNCs and top Cos. The Indian government has initiated a plan to recognize skill certifications awarded by top corporates, including multinationals, to their employees by standardizing and assigning credits to their skills training and bench marking it to international standards. The National Council for Vocational Education and Training (NCVET), under the government’s Skill India Mission driven by the Ministry of Skills Development and Entrepreneurship, would provide the recognition to these entities. Officials said government recognition would lend credibility to the certification given by private entities and enable greater mobility for employees. Read more here.
Britain - Issues over AI based Surveillance Technology The use of live facial recognition (LFR) technology — which creates bio metric facial signatures before instantaneously running them through a watch list of suspects — led to 10 arrests for crimes including threats to kill, bank fraud, theft and possession of a crossbow. The targeted use of AI for identifying such anti-socials and criminals has been welcomed. This is also in accordance with the AI laws being developed by the EU. Read more here. However, a study says that deployment of technology in public failed to meet ethical and human rights standards. Read more here.
More control over your location data - Google Maps As per the Google Blog, location history or ‘timeline’ data will soon be saved on the users device if it is turned on. It will be possible to delete all or part of your location history data at any time. When you’re changing your phone you can back up your data on the cloud and it will be encrypted. The auto delete function of timeline is being set to delete every three months whereas previously it was 18 months. Read more here.
A word of caution Coding and chat services are areas where AI based automation is happening surely and steadily. Despite the rhetoric, there can be no doubt that companies will pursue the need to increase their efficiencies and reduce their costs. They will resort to AI based automation and the use of products like Microsoft Co-Pilot and Bard. Employees working in these areas should focus on acquiring additional skills such as managing code development through the use of Co-pilot and domain knowledge required in the their company’s core business areas. It may not be required to transition to an area where AI has not yet made its presence felt, rather a good approach would be to find ways to work along with AI. The time for working hand in glove with AI has come.
AI Summary news 11 Dec 2023:
(Source Tech Crunch) Hype around the release of Google’s “Gemini” chatbot co-founded by Google DeepMind and Meta alumni has taken off. The release of Gemini Pro product is still awaited
Mistral closed one of Europe’s largest seed rounds to date prior to Friday’s 8th Dec 2023 fundraise of Euro 450 mn when it hasn’t even launched a product yet. Investors are probably viewing Mistral — as well as its sometime rival, Germany’s Aleph Alpha — as Europe’s opportunity to plant its flag in the fertile generative AI ground.
Meta and IBM teaming up for AI support “open innovation” and “open science” in AI.
OpenAI is also looking to set up a local team in India to help the AI startup navigate the Indian policy and regulatory landscape.
Governance:
The cozy relationship between OpenAI and Microsoft, a major backer and partner, is now the focus of a new inquiry launched by the Competition and Markets Authority in the U.K. The FTC in the US is also scrutinizing the Open AI – Microsoft relationship under a magnifying glass.
Lobbyists, led by Mistral, have in recent months pushed for a total regulatory carve-out for generative AI models. But EU lawmakers have so far resisted such an exemption. Read more here.
Europe enacts laws governing AI
The accord requires foundation models such as ChatGPT and general purpose AI systems to comply with transparency obligations before they are put on the market. Read more here.
AI news – Week 48
Here are some noteworthy happenings in the AI space in the last week.
Crowd sourced feedback to help train Robots:
In many instances, a human expert must carefully design a reward function, which is an incentive mechanism that gives the agent motivation to explore. The human expert must iteratively update that reward function as the agent explores and tries different actions. This can be time-consuming, inefficient, and difficult to scale up, especially when the task is complex and involves many steps. Researchers from MIT, Harvard University, and the University of Washington have developed a new reinforcement learning approach that doesn’t rely on an expertly designed reward function. Instead, it leverages crowdsourced feedback, gathered from many nonexpert users, to guide the agent as it learns to reach its goal. Read more here.
Public opinion being eroded by another AI threat
Elon Musk has integrated ‘X’ with ‘Grok’, his LLM which is being trained on the latest sentiments posted on X. Just imagine the confusion that could be created by this. There are sentiments expressed on X by all sorts of people at the extreme ends of the political, ideological, national and other spectra. There is no charge on posting on X, so those who post are trigger-happy and for many it sometimes becomes a slanging match and a way to pass time. Is there going to be a value add or value detraction? Will the voices of the public be drowned out by this?
Current AI news around the world
20th Nov 2023
Breakthrough approach by the EU?
Sourced from Reuters
A big breakthrough for regulation of AI seems to have happened. France, Germany and Italy have reached an agreement on AI regulation which is expected to accelerate negotiations to finalise the EU AI Act, a 3 tier risk classification model based on the level of risk posed to fundamental rights and user safety. “The model cards shall include the relevant information to understand the functioning of the model, its capabilities and its limits and will be based on best practises within the developers community “ the paper said. The approach is not to regulate AI itself, but rather its application, which does seem to make good sense. Read more about it here.
Older News Items
2. US government plans AI governance agenda
Federal departments have been tasked with drafting AI regulations. The FTC has been tasked to focus on “anti-competitive behaviour” and “consumer harm”. Read here.
3. Sunak’s AI Guest List Risks Falling Short as Leaders Hold Back