U(useful) AI?
I may lean towards AI-sceptic but I am not ant-AI. I believe AI has it's purposes. As long as we remember that it's just another tool. However, at the moment, it's like a Swiss Army knife, a Jack of all trades, master of none. How many times has a Swiss Army knife been used to remove a stone from a horse's hoof? Ok. I have used the corkscrew and the bottle opener on occasion when no alternatives were at hand, and the knife was always sharp. But it is a glorified pen knife, admittedly with extras that can be life saving in times of extreme urgency and need, but are, more generally, obstacles to overcome or attempt to ignore as you try to find the blade, did I mention how sharp it is? Slice the top of your finger off if you're not careful. Only safe one to extract is, of course, the corkscrew. However, the corkscrew has deep feelings for the cork and will ultimately choose to remain with the cork after breaking off with the Swiss Army knife.
Jack of all trades, Master of none.
I have used AI, ultimately successfully, to translate one codebase into another. I used it as a tool. I had a project I had written in Python and wanted a Cython version to compile into binaries. I could have done it myself, after all, I wrote or refactored most of the multi file, multi process, Python behemoth myself over 2 years, but I'm 58, I don't have that much time to waste. Also, it would have bored me to tears, meaning I would have made countless mistakes leading to depression, frustration and resentment. Not good. So I used AI to do it. I was lucky, some old grey beard had written or refactored most of the Python code. So it was modularised and de-coupled. Used relevantly named classes, functions, constants and variables. This was invaluable when diffing both versions side by side. Each AI generated Cython module or class could be called from the known good Python build. Any Cython errors were fed back to the AI and I was usually given the fix at the first time of asking. Occasionally 2 or 3 iterations were required. Very rarely, after the 3rd iteration failed, and when I could see the rabbit hole looming in front of me, I would step in. In those few interludes, scanning both codebases, I completed a crash course in Cython! Ultimately it compiled. Woohoo! (Un)fortunately, old grey beard also had a test plan from the original Python: Unit tests, thanks AI!; Integration tests, thanks grey beard!; User Acceptance test, thanks QA! So iterating on the test failures in the same way as previously done for the compilation errors, with the same ratio of outcomes (80% 1 iteration; 15% 2-3 iterations; 5% grey beard) and learnings, we finally had a complete, fully functional Cython codebase from which to compile complete, fully functional binaries. In a week!
Use the right tool for the specific job.
I believe in AI. I don't believe in AGI. Or, more pertinently, GAI, General Artificial Intelligence. GAI is what our main tech firms are trying to build. They are striving to be all the answers to everything. But what they really want are the questions. They want to know what you want, what you desire, what you fear, not what you need. They want to know your questions so that they can sell you their answers. I believe that AI, highly focused on narrow domains, trained on reliable, accurate data sets, has the potential to unlock previously hidden insights such as in: cancer research, virology, nuclear fusion, urban planning, habitat restoration, renewable energies and ... better weather predictions! Truly useful. Domain constrained, reliably accurate and learning from it's mistakes. That's what I want from AI. That AI tool will work. Right specific tool for the specific job. There should, and has to be, innumerable tailored AIs for the gamut of specific jobs it can help with.
I don't believe in AGI. Wrong tool for any and all jobs. I don't believe in AGI, because everyone focuses on the I, Intelligence, passing over the A, Artificial. It's not intelligent, in the same way as artificial leather is not real leather, as artificial limbs are not flesh and blood limbs. Artificial Intelligence is not intelligent. However, if trained on accurate and relevant information, and given strict parameters to lower the amount of random hallucinations, AI can be extremely useful.
The great thing about UAI, Useful Artificial Intelligence, is that you don't need to burn the planet to benefit from it. The heavy lifting has already been done. Every AI out there understands what you are trying to say, and every AI out there can explain it's findings in a way you can comprehend. And you can run a model that can do that on your laptop, a Raspberry Pi, or, even your phone. No Internet required. No cloud. No, "I agree my data may be used for training purposes". Your data, under your control. An AI truly your own*. Ready to listen, explain and help, and not tell anyone else anything. Add your own data, documents, spreadsheets, photos, emails, videos, etc to the model; add your own agents**, Nostalgic, Scrooge, Public Speaker, Work etc; point the agents to siloed sets of your data and sift results, there will be some wheat amongst the chaff. All on your personal devices, under your own control. The processing and data storage necessary for this highly focused, personalized assistant, is already readily available on your devices.
I believe that AI will be revolutionary. However, it does not need to be concentrated in the hands of Big Tech. AI will prove its worth when it is commodified and integrated into existing tools and processes. Not surplanting or replacing any individual part of the process or tool, but, rather, augmenting the tool or aiding the process to make it more (Artificially) Intelligent.
Follow up:
I then went on to Cythonize my Windows based - the original, server program, was on Linux - GUI into a compiled, single file executable. Leaning on lessons learned. Following the same iterations as before. I had a fully, tested working, complete, obfuscated executable in.... 2 days! AI's in my toolbox now.
GIGO always has been and always will be true. Most of the time that means Garbage In, Garbage Out. However, sometimes, I'm beginning to think, it can mean Good In, Good Out.
