7 ms·

> Deterministic if/then statements This was the pinnacle of AI in the 80's. They called them "expert systems".

klyrs 4y ago
Principle problem being that expert systems required meticulous inputs from domain experts, codified by skilled engineers. People don't have time or startup capital for actual expertise...
- 10000truths 4y ago
  And AI requires the same thing, we just call them data scientists and ML engineers. Using linear-ish algebra instead of decision trees doesn't change the the fact that you need time and capital to hire experts.
  - yccs27 4y ago
    The big difference is that data scientists only work on the model architecture and data sources, whereas expert systems need people who have expertise in the subject matter itself. One of the biggest changes from 'old AI' to modern ML is that we no longer try to use human domain knowledge as much, instead getting the model itself to see the same pattern from data.
    - y2kenny 4y ago
      Isn't labeling still an important part of modern AI?
      - daveguy 4y ago
        Yes, but there is a whole field of artificial intelligence called unsupervised learning that tries to identify labels without pre-defined labels. At the extreme end there are no externally imposed / defined labels and artificial labels are determined by empirical clusters or some orthogonal data pattern or algorithm. Unsupervised learning is much less effective and not as mature as supervised learning. In the case of LLMs the label is "next words" and it's inferred from a corpus of text.
      - yccs27 4y ago
        I'd say labels (for supervised ML) are fundamentally different from rules (for expert systems), because - labels are easy to decide in many cases - rules require humans to analyze patterns in the problem space - labels only concern each data point individually - rules generalize over a class of data points
- strangemonad 4y ago
  Which is very much unlike today’s training set inputs meticulously labeled by domain experts and curated by engineers?
kikokikokiko 4y ago
Large language models are the thing the average joe in 2023 would call AI the most, and at the end of the day, if you go deep enough down the 500 billion parameters rabbit hole, it's just a "veryyyyyyy loooooong chain of if-then-else's" obtained after 10s of thousands of hours of computing time over basically all of the text generated by humans over 30 years of internet existence. I know it's not EXACTLY that, but it could be pretty much "recreated" using this metaphorical long chain.
- onethought 4y ago
  This is kind of like saying every video could be recreated manually with painters. That is, this is a ridiculous statement.

Principle problem being that expert systems required meticulous inputs from domain experts, codified by skilled engineers. People don't have time or startup capital for actual expertise...

10000truths 4y ago
And AI requires the same thing, we just call them data scientists and ML engineers. Using linear-ish algebra instead of decision trees doesn't change the the fact that you need time and capital to hire experts.
- yccs27 4y ago
  The big difference is that data scientists only work on the model architecture and data sources, whereas expert systems need people who have expertise in the subject matter itself. One of the biggest changes from 'old AI' to modern ML is that we no longer try to use human domain knowledge as much, instead getting the model itself to see the same pattern from data.
  - y2kenny 4y ago
    Isn't labeling still an important part of modern AI?
    - daveguy 4y ago
      Yes, but there is a whole field of artificial intelligence called unsupervised learning that tries to identify labels without pre-defined labels. At the extreme end there are no externally imposed / defined labels and artificial labels are determined by empirical clusters or some orthogonal data pattern or algorithm. Unsupervised learning is much less effective and not as mature as supervised learning. In the case of LLMs the label is "next words" and it's inferred from a corpus of text.
    - yccs27 4y ago
      I'd say labels (for supervised ML) are fundamentally different from rules (for expert systems), because - labels are easy to decide in many cases - rules require humans to analyze patterns in the problem space - labels only concern each data point individually - rules generalize over a class of data points
strangemonad 4y ago
Which is very much unlike today’s training set inputs meticulously labeled by domain experts and curated by engineers?

kikokikokiko 4y ago

Large language models are the thing the average joe in 2023 would call AI the most, and at the end of the day, if you go deep enough down the 500 billion parameters rabbit hole, it's just a "veryyyyyyy loooooong chain of if-then-else's" obtained after 10s of thousands of hours of computing time over basically all of the text generated by humans over 30 years of internet existence. I know it's not EXACTLY that, but it could be pretty much "recreated" using this metaphorical long chain.

onethought 4y ago
This is kind of like saying every video could be recreated manually with painters. That is, this is a ridiculous statement.