Q* and How It Might Impact Society 🤖
Q* and How It Might Impact Society 🤖
Prior to Sam Altman’s firing as OpenAI CEO, several staff sent the board a letter warning of a development they said could genuinely threaten humanity - it's known as Q*.
Q* (pronounced Q-star and not to be confused with Amazon’s new Q chatbot), is reportedly OpenAI’s latest machine-learning model. Information about it remains scarce but its capabilities have caused concern amongst staff.
It’s even been suggested that its development was a large part of the reason for Sam Altman’s removal as CEO by the OpenAI board 2 weeks ago.
What Is It? 🤔
The name suggests that Q* could be a combination of two well-known AI algorithms - Q-learning and A* Search.
Q-learning is a form of reinforcement learning, a technique in which AI acquires decision-making skills through a process of trial and error.
In this approach, a neural network is trained to work out the best action to take by assessing the quality of different combinations of actions and states.
A* search is a pathfinding algorithm, which efficiently finds the shortest path between nodes in a graph. It estimates the lowest cost from a node to the goal, combining this with the cost to reach that node.
The combination of these two approaches means that, unlike ChatGPT’s current RLHF (Reinforcement Learning through Human Feedback) approach, the new approach does not rely on humans to learn; it learns on its own.
If true, it would mean that the algorithm could learn to understand the world without the need for human input.
Since OpenAI has yet to comment on the speculation publicly, for now, it remains just that. But that hasn’t stopped staff inside the company from voicing their concerns…
AI Concerns ⚠️
Although it sounds like something out of a sci-fi movie, according to Reuters, several staff wrote to the board warning of an AI discovery that they believe could threaten humanity.
People on the OpenAI community forums certainly seem concerned about both the breakthrough itself and OpenAI’s apparent unwillingness to share more about it.
The model was apparently demonstrated internally and is capable of solving simple math problems.
This might not seem like much but it would mean that the AI is capable of solving problems unrelated to the data it has been trained on - a huge leap forward from the current large language models.
"If it has the ability to logically reason and reason about abstract concepts, which right now is what it really struggles with, that's a pretty tremendous leap" said Charles Higgins, co-founder of AI-training startup, Tromero.
Potential Impact 🤖
There’s still a lack of concrete info online about the new algorithm and OpenAI hasn’t formally announced whether it exists, but the impact of Q* (if/when it’s released) will likely represent a huge leap in AI system capability.
The fact it can learn and adapt itself autonomously in changing and complex environments means it could be applied to a whole new set of problems.
Advances could be made in everything from self-driving, to walking and physical manipulation and it would mean even more rapid progress in areas like science, engineering and automation.
Of course, this also comes with the normal concerns related to AI development. These include the economic impact and level of job displacement that a huge machine learning breakthrough could create and ethical and alignment issues.
It’s unlikely that Q* really does represent a huge step towards Artificial General Intelligence (AGI), but even if it brings us a bit closer, it will still have a huge impact on modern society.