4 ms·
99% of the 'AI' startups, because the 'AI' is usually team of humans somewhere in India. Fake it till you make it, I think? I'm not saying this to be negative,
by LeonM 8y ago
99% of the 'AI' startups, because the 'AI' is usually team of humans somewhere in India. Fake it till you make it, I think?
I'm not saying this to be negative, as this will actually work to validate product-market-fit before investing in the AI development.
Most of these companies will fail though, as their pricing will be based on a 100% AI solution, which is rarely achievable.
- disqard 8y agoI'm not trying to undermine you, but can you share a link or source? I too, have a vague feeling that many "AI Startups" are trying to ride the buzzword train, and actually doing their work using humans. However, I haven't read any good journalism actually showing that this is the case.
- griffinkelly 8y agox.ai: https://www.bloomberg.com/news/articles/2016-04-18/the-humans-hiding-behind-the-chatbots https://www.bloomberg.com/news/articles/2016-04-18/the-human...
- dullroar 8y agoAnd this: https://www.bbc.co.uk/news/technology-46055595 https://www.bbc.co.uk/news/technology-46055595?
- nsenifty 8y agoThat's just getting data for training isn't it? You can't train AI systems by AI.
- dentemple 8y agoSounds like a new startup idea to me
- newshost 8y agohttps://cloud.google.com/automl/ https://cloud.google.com/automl/
- gumby 8y agoNIs train NIs, so in principle why not? (yes, we're a long way from there today but fixing that is what I'm working on)
- derefr 8y agoA human baby can't learn to walk by having it demonstrated to them; they actually have to try walking themselves. (Though maybe this would be different if adults were shaped exactly like babies, and babies could understand speech well enough to parse a description of the process of walking.) Heck, even after we learn to talk and read, we still don't learn to write by having it explained, and only vaguely learn by demonstration. We mostly just learn by generating outputs and having them evaluated by the other NI (sort of like how AlphaGo Zero works.)
- disqard 8y agoI just saw this and came back to HN to share it :) Thanks!
- disqard 8y agoThank you for sharing this.
- ninjakeyboard 8y agoalso https://arstechnica.com/information-technology/2017/11/expensify-acknowledges-potential-privacy-problem-by-calling-it-a-feature/ https://arstechnica.com/information-technology/2017/11/expen...
- rcavezza 8y agoThis is 100% true. So, so, so true. In actuality, A.I. will not ever 100% solve all problems and some sort of human intervention is typically needed to serve as a QA purpose. AI can get you 80%, 90%, 95%, and 99% of the way to solve a problem. When you first start out, a human element will be doing most of the heavy lifting and creating training data for you. As you iterate on the model and improve it, the human intervention element will be less and less time consuming over time. As you near 100% accuracy, you will probably still want a human intervention element acting as QA because all models have some rate of error.
- alexcnwy 8y agoYou can use the model's output probabilities to optimize QA routing - any cases with less than 90% confidence to humans but save humans doing the cases that can easily be done by models :)
- bendergarcia 8y agoAI: Artisanal Intelligence
- nostrademons 8y agoI like to call it "aggregated intelligence", because that's what basically all AI solutions really are. Even ones that are actually using embeddings & CNNs are just aggregating the opinions of the thousands to millions of humans who supplied training data.
- derefr 8y agoThat's a good name. It contrasts pretty well with the 1970s' paradigm of "expert systems"—where you were encoding the opinions of one human, or the consensus opinion of a conference room full of humans.
- andysinclair 8y agoI like “Artificial Artificial Intelligence”.
- antaviana 8y agoThat reminds me that in an AWS conference a few years ago, the presenter touted Amazon Mechanical Turk as "Artificial Artificial Intelligence"
- nradov 8y agoThere are so many of those fake AI startups now that you could probably build a successful startup specifically to provide offshore human staffing services and workflow automation to them. During a gold rush, sell shovels.
- Qworg 8y agoThere are a few of these already - Mechanical Turk is the most famous.
- derefr 8y ago...and then lower your costs by building a real AGI to pretend to be humans pretending to be AI. (Has anyone done this particular sci-fi plot?)
- spaceknarf 8y agoYes, in the book Life 3.0 by Max Tegmark! (An AGI is implemented that makes money on Amazon Mechanical Turk.)
- Havoc 8y agoThere was an article on here like two days ago about a company in Kenya that does training sets for big tech. 1000s of people marking cars on pictures etc
- puranjay 8y agoI pitched a local language startup idea to a few companies. I was summarily tolf that it didn't have any "zing" to it. So I added "AI" in all my pitches and got a lot of interest. Didn't pursue the idea further because I didn't really believe in it, but it just proved to me that investors often chase buzzwords instead of ideas
- nautical 8y agohttps://testzap.com https://testzap.com does exactly that for app testing, although there are few products in market claiming to do AI based testing.
- the_other_guy 8y agoseriously how do people get away with such things legally? this is total fraud