9 ms·
> Real Risk: Overreliance on inaccurate tools We might be able to remediate this somewhat by designing AIs that are purposely below some threshold of reliabili
by hirundo 4y ago
> Real Risk: Overreliance on inaccurate tools
We might be able to remediate this somewhat by designing AIs that are purposely below some threshold of reliability. Their current tendency to hallucinate will discourage over reliance if we can't or don't fix it.
It's a little like the distinctive smell that is added to propane to make it easier to detect leaks. By adding, or not removing, easily detectable hallucinations from AI, it's easier to detect that the source must be checked.
We desperately want oracles and will quickly latch on to highly unreliable sources at the drop of a hat, to judge by various religious, political and economic trends. It's inevitable that many will turn AIs into authorities rather than tools as soon as they can justify it. We could delay that by making it less justifiable.
- nsajko 4y ago> We desperately want oracles and will quickly latch on to highly unreliable sources at the drop of a hat, to judge by various religious, political and economic trends. It's inevitable that many will turn AIs into authorities rather than tools as soon as they can justify it. Indeed. This was true even for the primitive AI/chatbots that existed in the sixties: https://en.wikipedia.org/wiki/ELIZA_effect#Origin https://en.wikipedia.org/wiki/ELIZA_effect#Origin
- CatWChainsaw 4y agoIt's a little funny, isn't it, that physical products have planned obsolescence to make them fall apart faster, to get people to buy a newer model with only small changes. Yet I can't see an "intellectually stunted" flying as a marketing tool precisely because it's not the best "hire".