My timeline for AGI hasn’t changed much, but my timeline for ‘semi-transformational narrow AI’ has become shorter.
Lots of academics are talking about how ChatGPT (and student and teacher use thereof) will force a revolution in how we assign homework, writing assignments, exam questions, and even class discussion questions. The whole experience of school and college will either have to change dramatically (e.g. a return to in-person lectures & discussions, and in-person paper-and-pencil exams), or schools and colleges will become empty rituals in which teachers using AIs pretend to teach and grade things, and students using AIs pretend to learn things.
Similar pressures will arise in many other industries, social practices, and relationships.
I think what ChatGPT highlights, fundamentally, is that even ‘narrow AI’ will be transformational enough to impose a dizzying rate of change on our civilization, and to impose qualitatively new kinds of risks.
This sounds a bit exaggerated. I haven’t engaged much with ChatGPT, but it seems like the same arguments could be made for students having access to the internet. Can you define what kind of writing assignments you are referring to?
Konstantin—For example, if I’m running an online discussion forum in one of my psychology classes, and one student has a question (e.g. ‘What is the difference between classical and operant conditioning?’), another student could answer by simply typing that question into Chat GPT, copying the answer into the Blackboard discussion forum, modifying it a little bit to defeat any anti-plagiarism software, and getting credit for the answer.
Of, if an online exam asks for a short essay answer to a question (e.g. ‘What are some limitations of evolutionary psychology?’), a student could get a pretty good answer from Chat GPT, copy it into the online exam software, and get a good grade.
Or, if a student is developing longer term paper on some topic, they could ask Chat GPT ‘Please write an outline of a term paper on how multivariate genetics can inform psychiatric diagnosis’, and they could easily build upon that.
If you try these three example prompts, you’ll see that Chat GPT does an amazingly good job of doing about 80% of the students’ work for them, much faster than any student could write.
I used chatgpt and it is amazing good at telling historical stories and describing people. I have used a similar tool named Jasper and is equally as powerful.
The data we are generating now from these systems are just mind boggling, Elon was right from years before—it will come a time that the data generated by AI systems will dwarf human outputs at scale. We are seeing the early stages of it.
Thank you. I think for the first two I still don’t see a big jump from just internet access. But I agree that cheating got a lot easier. The third example seems particularly like outsourcing an important skill.
I think my skepticism comes from the fact that you could already easily cheat on my biology exams with internet access (and people did when they were online during covid). This isn’t much of a problem with pen & paper exams, but it seems that online degrees will really struggle now.
Out of curiosity I fed it some questions from assignments I had during my undergrad. Predictably, it got answers asking for simple factual knowledge quite right. It also did quite well in delivering a bone structure for essays/reports.
But it completely failed on more complex questions related to maths or algorithms. The answers still read really well, they were just completely false.
My best guess is that we will soon all have ai assistants, which will certainly shift teaching and being taught, but I don’t think we have to panic yet.
Anton—this sounds accurate. Chat GPT seems much stronger on qualitative verbal answers than on math, algorithms, and logic.
But, outside hard-core STEM fields, a lot of college courses focus on qualitative verbal answers. So, those of us who teach in the social and behavioral sciences will face stronger challenges than those who teach in computer science, physics, or statistics.
My timeline for AGI hasn’t changed much, but my timeline for ‘semi-transformational narrow AI’ has become shorter.
Lots of academics are talking about how ChatGPT (and student and teacher use thereof) will force a revolution in how we assign homework, writing assignments, exam questions, and even class discussion questions. The whole experience of school and college will either have to change dramatically (e.g. a return to in-person lectures & discussions, and in-person paper-and-pencil exams), or schools and colleges will become empty rituals in which teachers using AIs pretend to teach and grade things, and students using AIs pretend to learn things.
Similar pressures will arise in many other industries, social practices, and relationships.
I think what ChatGPT highlights, fundamentally, is that even ‘narrow AI’ will be transformational enough to impose a dizzying rate of change on our civilization, and to impose qualitatively new kinds of risks.
This sounds a bit exaggerated. I haven’t engaged much with ChatGPT, but it seems like the same arguments could be made for students having access to the internet.
Can you define what kind of writing assignments you are referring to?
Konstantin—For example, if I’m running an online discussion forum in one of my psychology classes, and one student has a question (e.g. ‘What is the difference between classical and operant conditioning?’), another student could answer by simply typing that question into Chat GPT, copying the answer into the Blackboard discussion forum, modifying it a little bit to defeat any anti-plagiarism software, and getting credit for the answer.
Of, if an online exam asks for a short essay answer to a question (e.g. ‘What are some limitations of evolutionary psychology?’), a student could get a pretty good answer from Chat GPT, copy it into the online exam software, and get a good grade.
Or, if a student is developing longer term paper on some topic, they could ask Chat GPT ‘Please write an outline of a term paper on how multivariate genetics can inform psychiatric diagnosis’, and they could easily build upon that.
If you try these three example prompts, you’ll see that Chat GPT does an amazingly good job of doing about 80% of the students’ work for them, much faster than any student could write.
I used chatgpt and it is amazing good at telling historical stories and describing people. I have used a similar tool named Jasper and is equally as powerful.
The data we are generating now from these systems are just mind boggling, Elon was right from years before—it will come a time that the data generated by AI systems will dwarf human outputs at scale. We are seeing the early stages of it.
Thank you. I think for the first two I still don’t see a big jump from just internet access. But I agree that cheating got a lot easier. The third example seems particularly like outsourcing an important skill.
I think my skepticism comes from the fact that you could already easily cheat on my biology exams with internet access (and people did when they were online during covid). This isn’t much of a problem with pen & paper exams, but it seems that online degrees will really struggle now.
Out of curiosity I fed it some questions from assignments I had during my undergrad. Predictably, it got answers asking for simple factual knowledge quite right. It also did quite well in delivering a bone structure for essays/reports. But it completely failed on more complex questions related to maths or algorithms. The answers still read really well, they were just completely false. My best guess is that we will soon all have ai assistants, which will certainly shift teaching and being taught, but I don’t think we have to panic yet.
Anton—this sounds accurate. Chat GPT seems much stronger on qualitative verbal answers than on math, algorithms, and logic.
But, outside hard-core STEM fields, a lot of college courses focus on qualitative verbal answers. So, those of us who teach in the social and behavioral sciences will face stronger challenges than those who teach in computer science, physics, or statistics.