A PhD-Level AI Agent

Of course, when I learned that OpenAI has been considering launching the PhD Level Agent at $20,000 a month, I thought, "Hey, that would be a nice salary to have for a PhD graduate like myself". After all, AI has been promoted as cheaper than humans, which doesn't sound correct for doctorate holders - FYI. But seriously:

What is a PhD?

It is NOT experience - or at least not just. Spending 10 years doing the same thing over and over again or being led by your leaders' vision is not a PhD-level experience. In certain areas, there are individuals who, in my honest opinion, merit a PhD based on experience.

Let me explain. When you enrol in a Bachelor's or Master's degree, you have modules and exams per semester. Assignments are short, exams are focused. In other words, you have someone to guide your direction. When you work, you are guided by metrics, priorities and business requirements.

When you enrol for a doctorate degree, you aim to solve a problem. You start by being guided, but you end up exploring avenues and methods in your own way. Which is why most PhD graduates conflict with their supervisors - different ideas, different thinking. At work, you can have different ideas, but you need to prioritise them based on your business requirements. You are led by your leaders, your performance is measured by actual deliverables, and you have a company of individuals for support. You may not agree with the direction, but you do what is asked because your focus is to support the business in reaching its objectives. Moreover, you have financial stability, whereas PhD candidates are either on a scholarship or working part-time jobs.

As a PhD candidate, you have a supervisor to brainstorm with, but you are expected to lead the research, build your hypotheses, develop and execute your experiments, evaluate the results, and when it fails, start again, most times from scratch.

Sounds familiar? Should be, this is what Agentic AI is simulating!

You can't shrug off the fact that your manager didn't have a good idea or guided you in the wrong direction, because you're your manager. You can't even blame a colleague because you're that colleague. And my favourite:

You can't always blame the data - at least not in astrophysics. I couldn't ask a galaxy to move to a part of the Universe where the SNR is favourable, for obvious reasons.

So when OpenAI speaks about PhD-Level agents, in my opinion, it is referring to the skill of managing your own work, thinking outside the box, finding a solution and keeping failing until you succeed. It refers to providing these agents the ability to figure it out on their own.

Essentially, they're promoting a PhD Level of resilience, endurance and perseverance. When pursuing a PhD, saying "No" to an idea you don't like without first considering it shows inexperience and a lack of intellectual maturity. We all know that this is different in the workplace. A PhD candidate tests all avenues, even the improbable ones, and turns a pencil into the smartest calligraphy pen there ever was, because asking for more resources is not always feasible. At work, you put in that ticket request, and it is handled. A PhD candidate must submit a funding proposal for a specific project.

Ask yourself, what do you say when someone proposes something you don't understand or dislike? "Let's chat" or outright "No"? Or worse, you come up with reasons why that won't work? Were you able to get away without knowing the details? Hint: A PhD viva voce lasts 3 hours (at least mine did) - you cannot not know! I may be wrong, but I think what OpenAI is promoting by a PhD-level agent isn't expertise in a specific area. Rather, they're promoting the behaviour of a PhD graduate: to take on a topic, knowing nothing or very little, to acquire knowledge and skills, go deep into its core, and persevere. Because let's face it, every topic can be taught, but the same cannot be said for resilience and endurance.

If it were, everyone would do a PhD.