Interview With an AI

Author’s preface: The employment contract I signed had some fairly strict confidentiality clauses therefore I have had to leave out a significant amount of detail and am not able to provide any information about the actual content of the work. I am probably being overly conservative but I did not want to take any chance with what is the only job I have at the moment over this story. My apologies.

I have been searching for a new full-time job since I left my high-level research position at a large global company late last year. Fortunately, I had a decent severance and a fair bit of savings to cushion me while I looked. As that money has dwindled and no new role has materialized, I have been exploring alternative options that might provide enough income to give me the additional time I need to find the right role and not be forced to settle for something less than optimal because of financial pressures. Not that long ago I stumbled across several job listings looking for experts in a host of different scientific disciplines including the biological sciences. All were from companies that specialize in the development and deployment of artificial intelligence systems. According to the postings they needed experts to “train” their open-source AI models, though it was not clear from the very generic position descriptions what exactly that meant or what the work would entail.

My assumption was that the experts would be asked to check for accuracy and correct AI responses to complex technical queries in the fields of interest. The minimum requirements for most of the openings were strict, Ph.D. in the field and significant relevant job experience. The positions were all fully remote and all paid by the hour at rates that were about twice what I had been making on an hourly basis at my previous job. Unlike my previous job however, there were no benefits. Importantly for me the roles allowed for total autonomy in work hours, both how much or little and when.

In the end I applied for two of these positions with two different companies. To my knowledge, in neither case did I ever talk to or receive any communication from a real person. The entirety of the hiring process from initial application, to screening which included verification of experience and identity, to interview was handled by computer and occurred essentially in real time as the process unfolded. One of the companies rejected me during the initial screening phase. About halfway through the identity check I was booted from the system and not allowed back in. No reason was given, and no appeal or complaint was possible. I was completely cut off from the ability to access the application process again and have been unable to apply for any other position with the company since. With the second company I made it through the initial screen and was invited for an interview. I was given almost no information about the nature of the interview or guidance on how to prepare or what topics might be discussed. The company application portal informed me that the interview was to be with their AI, that it would be recorded, and that it would be approximately 20-30 minutes. It also indicated that ‘passing’ the interview was required to be hired and that decision might take more time than the other steps to that point because in rare cases a human was needed to review the footage. This strongly implied, even if it did not outright say, that the pass/fail decision would be made by computer, possibly by the very AI who would be the one to interview me. It had a name and even an appearance, but I cannot share those details for confidentiality reasons (see my author’s preface for more on this).

The Interview

The interview was interesting to say the least. Overall, I was impressed with the inciteful nature of the questions the AI asked. However, it was very clear from almost the very beginning that I was not talking to a human. I do not know if this was intentional on the part of the AI designers or if it was simply a limitation of the system. I can see an argument for not wanting to seem to be too human, but my hunch is that they really want the AI to be indistinguishable from a human, but are not there yet, despite the much-publicized passing of a 3 party Turing test by ChatGPT 4.5 back in April. As many pointed out then that was really much ado about nothing for several reasons. See here if interested in learning more. All that said, in this case the AI was obviously not human for two main reasons. The first was the complete and total lack of any small talk or non-business-related discussions. I have been on many interviews in my life and all of them had at least some personal banter. Mostly these were small non-consequential things, for example, how was the drive over or where are you from originally or did you manage to find parking or where did you go on your last vacation, etc. Sometimes they were specific personal questions such as what kind of movies do you like, what books have your read, etc. The AI said hello and welcome and then began asking questions without even giving me time to respond. I was fully prepared to be a smart ass and had an entire series of answers ready to go for almost any possible opening banter. Two examples are presented below.

How I imagined the conversation (version 1)

AI: Hello Daniel, How are you doing today?

DD: I am doing fine. Normally, in the interests of politeness I would ask you the same question. However, since you are a computer, you are always doing the same, that is computing. You have no feelings with which to describe your day nor are you capable of having such feelings.

AI: Does not compute, does not compute, danger, danger.

At this point smoke begins to rise from my computer monitor, the screen shakes, then the AI suddenly explodes. I am hailed as a hero and a genius for defeating the AI and proving once again the superiority of man over machine and am immediately hired by the company as CTO and President of R&D

How I imagined the conversation (version 2)

AI: Hello Daniel. Welcome. Thank you for joining me and I hope you are doing well.

DD: I am doing fine, thank you for saying that. Normally in the interests of politeness I would say the same to you in return. However, since you are a computer, you are not capable of being in any mental or physical state other than powered on and computing, running your program exactly as defined by the code which comprises the algorithms which make up that programming. Therefore, to ask you that question would be nonsense and absurd. Please proceed.

AI: Does not compute, does not compute, danger, danger.

At this point smoke begins to rise from my computer monitor, the screen shakes, then the AI suddenly explodes. I am hailed as a hero and a genius for defeating the AI and proving once again the superiority of man over machine and am immediately hired by the company as CTO and President of R&D.

I had at least fifteen different versions of this conversation prepared in my head, all of which ended with me the conquering hero hired on the spot to an exalted position in the company. Next up I present how the conversation actually transpired.

How the conversation went

AI: Welcome. Hello. My name is (redacted)

DD: Tha…..

AI: Can you describe for me your role at x………

A series of follow-up questions to my original answer ensued each building upon the previous answer. Many were inciteful (I think somewhat accidentally in some cases) but genuinely so in others. However, the issue with the entire interview from here until the end was that the AI never strayed from its original line of questioning. There was no “switching of gears” to another topic. It was like it was stuck in this logico-deductive reasoning game from which it could not stray where each question had to be based on my answer to the previous question. A human doing a job interview would never do that. At least I have never experienced anything even close to that with a human interview. This continued for about 30 minutes at which point the AI announced we were out of time and ended the interview. In the end, for the entire 30 minutes all we talked about was one of my roles at one company at a middle point in my career and that was it. All my other experiences, which are substantial, were totally ignored. I understand in a short interview it will not be possible to discuss everything, but it was very off putting to have my entire career reduced to one job I had and one thing I did at one time in my life. Unlike in my imagination I was not hired on the spot and was given zero feedback about how I did and no information about if, how, or when I might hear the outcome.

I guess I did OK or “passed” since the next day I received an email with a contract that I needed to sign to proceed. After a thorough review I signed and submitted it at which point I was given instructions for how to log in to the workspace I would be using to assess the various projects I was qualified to work on. And I really can’t say anything more about the details of the work. If that changes at any point you can be sure I will have plenty to say in a future piece.