Has ChatGPTo1 just become a 'Critical Thinker'?

Critical thinking was a famous 21st century skill, that everyone thought AI could never solve. It just has. We are only just beginning to glimpse what many predicted -  that AGI is getting closer. 

What we have just witnessed is nothing less than a cognitive breakthrough from text generation to reasoning. Note that not all problems and tasks are to do with pure reason but the charge made against GenAI was that it could not plan and reason. That door has been blown wide open.

New scaling model

But what really matters is that this opens up a new future for AI, as it can be scaled. It was thought that we had hit a ceiling on scaling or at least little gains for huge scaling. It may well be that OpenAI have hit a ceiling on model scaling and have switched to this approach - we don't know. Inference (reasoning) can now be scaled, opening up a new world of real intelligence. Sure, there will be glitches, but that’s the point of releasing these models. We will now get hundreds of thousands of real use cases in the real world. The old days of release a perfect product are gone

Beyond the chatbot

This shift has moved Open AI beyond chatbots to agents. You’re as likely to ask it to DO something, than answer a simple question. It can do maths and science but just as importantly it can do critical professional tasks and projects, the bread and butter of the professions, whether it is health, law, finance, HR or L&D. So many more use cases are opened up by this functionality as it moves from single case or linear queries to solving complex problems.

Critical thinking

Critical thinking was one of the famous 21st century skills, that everyone thought AI could never solve. If we see Critical thinking as a reasoned and critical approach to solving problems, it is here. In fact, ChatGPT o1's performance in maths and reasoning is extraordinary. It literally goes away to think, then comes back with potential solutions. Remember that this model is doing fantastically advanced maths at the level of the Maths Olympiiad and problem solving on quantitative tasks. The same will be true of qualitative tasks and more commonly, a mixture of both.

At a more specific level, these new models can determine an optimised L&D intervention based on information provided about the goals, learners, learning and resources. It would avoid the all-too-common solution of slabbing out a course.

Coworker

They can predict and plan the stages of projects, share those plans for comment to allow projects to develop. In more qualitative tasks, such as those one finds in the real world of organisations, domain knowledge also matters. This is where things will develop quickly.

You have to see this new approach as not providing simple solutions to single prompts but predicting and planning, multi-stage tasks, with far more penetrative judgement. You can get it to do the market research or needs analysis, then scenario analysis to evaluate potential outcomes, then a risk assessment. It will also surface any uncertainties and risks as decisions are made across the project, using risks analysis to manage the project. Users will still be able to steer things but in an insightful and informed manner. 

A typical problem may be the introduction of AI into workplace learning. One would want to consider all the critical and non-critical learning tasks, choices available on learning delivery, based on the types of learning, nature of the learners, geographic and language spread and available resources. On top of this regulatory and legal issues would also have to be considered along with available resources, perhaps an estimated budget. This is the sort of project that will be executed by these new models with caveats along the way.

If one were to plan a course in education or in the workplace it could do an excellent job in planning an optimised approach that considered the critical points, tasks and pedagogic approach.

Skills assessment will be so much better, as it is a complex tasks to identify skills, skill gaps and solutions to the organisation’s needs. 

Automation

You can see how these agents will massively improve customer support, actually focussing and solving customer problems, not working to limited scripts. Any by the way the translation capabilities have also been improved. Performance support will be SO much better as, when you get stuck, it will solve your problem in the workflow. These systems will give more targeted and realistic solutions to your need at that moment.

Reason

Reason is not everything and it almost always has a context and involves domain knowledge in the real world. That's why the coworker approach is likely for the foreseeable future. There may also be limits to the Chain of Thought approach. We do not yet know much about costs, context window, limits on the 'thinking' process. What we do know is that this is a breakthrough. There will be others.

Conclusion

Many of the old criticisms have been dramatically reduced. OpenAI's o1 model is now being put through its paces in professional service where reflection and reason really matter, before coming back to you. It goes into its own private chain of thought. This means that the longer it thinks, the more likely it is to be better on reasoning tasks. 

Everyone rushing to bin their AI slides -  Can't do maths? Can't plan? Can't reason? How many Rs in strawberry? Limitations being eliminated week by week. This is a new paradigm for AI.

We are only just beginning to glimpse what many predicted that AGI is getting very close. This is one step for mankind but a giant leap for AI.