- commit
- ed0b015e92df55a0ff28d086a0ed520f89a19036
- parent
- 77df8c83f01a1f5532d89168c9fd1cfcb1d5e58a
- Author
- Tobias Bengfort <tobias.bengfort@posteo.de>
- Date
- 2024-02-04 14:09
post: ai
Diffstat
| A | _content/posts/2024-02-04-7-thoughts-on-ai/index.md | 111 | ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ |
1 files changed, 111 insertions, 0 deletions
diff --git a/_content/posts/2024-02-04-7-thoughts-on-ai/index.md b/_content/posts/2024-02-04-7-thoughts-on-ai/index.md
@@ -0,0 +1,111 @@ -1 1 --- -1 2 title: "7 thoughts ideas on AI" -1 3 date: 2024-02-24 -1 4 tags: [code, philosophy] -1 5 --- -1 6 -1 7 I am really not sure if I should write another post about AI. I mean, what's -1 8 the point? Everything has already been said, hasn't it? So here are my 7 -1 9 (probably boring) thoughts: -1 10 -1 11 ## 1. AI is a useful tool -1 12 -1 13 Whether you think the current form of AI deserves the label "intelligent" or -1 14 not, no one can deny that it solves a huge class of problems that previously -1 15 could not be solved by computers. I especially see improvements in -1 16 accessibility, where natural language processing, image recognition, or -1 17 summarizing information are extremely important. -1 18 -1 19 ## 2. AI is post-modern -1 20 -1 21 So AI is useful, but it is also deeply non-rational. Its whole point is that we -1 22 do not understand how it finds its answers and which kinds of biases are -1 23 involved. We see that the rational approach that defined modernity has failed, -1 24 but we do not yet have a framework to evaluate the non-rational approach. This -1 25 makes AI a prime example of post-modernity. -1 26 -1 27 ## 3. AI is stupid, but humans are stupid, too -1 28 -1 29 Some people keep telling us of all the flaws and biases in AI. The most -1 30 prominent (but untrue) example being the [neural net tank -1 31 legend](https://gwern.net/tank). I do think that this work is important, -1 32 because too many people blindly trust what a computer says. But our response -1 33 should not just be "we have to keep a human in the loop". -1 34 -1 35 Humans are also just statistical machines with flaws and biases. Our concept of -1 36 morality is inconsistent at best, and most of us cannot even confidently -1 37 perform simple math. Instead of relying on humans, we should find robust ways -1 38 to deal with those flaws. -1 39 -1 40 ## 4. AI is flawed in unexpected ways -1 41 -1 42 The one thing humans have going for them is that we are used to their flaws. -1 43 All of our systems are designed to limit the impact of human error. When AI has -1 44 a flaw, chances are that the flaw manifests in a way we are not prepared for. -1 45 -1 46 ## 5. Talking to AI is a weird way of talking to humans -1 47 -1 48 I keep coming back to the [2017 34C3 keynote by Charles -1 49 Stross](https://www.antipope.org/charlie/blog-static/2018/01/dude-you-broke-the-future.html), -1 50 in which he argues that our AI overlords are already here, and they are called -1 51 corporations: -1 52 -1 53 > they're clearly artificial, but legally they're people. They have goals, and -1 54 > operate in pursuit of these goals. […] Finally, our legal environment today -1 55 > has been tailored for the convenience of corporate persons, rather than human -1 56 > persons -1 57 -1 58 Talking to a corporations, e.g. by calling a support hotline, is weird. You are -1 59 talking to a human on the other end, but you understand that talking to that -1 60 person as a representative for a corporation is different from talking to that -1 61 person in private. They are not free to say what they think. They have to be -1 62 polite. And they are replaceable: if you call that same hotline again, you will -1 63 talk to the same corporation, but to a different human. -1 64 -1 65 This is a weird form of communication, but one we have adapted to. It is also -1 66 not the only one: -1 67 -1 68 - Reading a book is also pretty weird if you think about it. It is a one-sided -1 69 discussion with a person that might already be dead. Still, some people have -1 70 deep discussions with the bible or something. -1 71 - Reports from people who have used a phone for the first time suggest that -1 72 talking to a disembodied voice is weird. -1 73 - Even [talking in groups](https://doi.org/10.1016/j.copsyc.2019.06.032) is -1 74 weird. Humans have created formal systems to mediate who should speak next -1 75 and how much airtime everyone gets. Reactions to your statement only come -1 76 once people in the queue have gotten their turn, and by then everyone might -1 77 have forgotten what you said. -1 78 -1 79 Large language models (the kind of AI tech we are currently talking about) are -1 80 hive minds, made up from millions of texts written by humans. Talking to AI -1 81 is, in some way, a weird way of talking to those people all at once. -1 82 -1 83 ## 6. Being nice is always a good idea -1 84 -1 85 Should you say "please" when talking to AI? -1 86 -1 87 I say you should. Mostly to keep the habit. If we don't, I see future -1 88 generations having a hard time socialising because they are used to shouting -1 89 commands at their virtual assistants all day long. I also don't see any reason -1 90 not to. So I will keep trying to be nice in conversations, no matter with whom. -1 91 (It actually bugs me that most programming is expressed in the form of -1 92 commands.) -1 93 -1 94 ## 7. AI is alien -1 95 -1 96 To me, fear of AI feels awfully similar to a general fear of the "other", -1 97 whether that "other" is *real* AI, an alien life form, or simply someone from a -1 98 different country. What reason would they have to attack us? Why should we even -1 99 make a distinction between "us" and "them"? -1 100 -1 101 I am not saying that we should be naive. But it's worth examining whether our -1 102 suspicions are rooted in valid concerns or if they stem from our tendency to -1 103 fear the unknown. -1 104 -1 105 ## Conclusion -1 106 -1 107 Here is a conclusion that ChatGPT wrote for this article. I actually like it! -1 108 -1 109 > So there you have it. I hope that you've found a nugget or two of interest. -1 110 > And if not, well, at least we've shared a moment of honest boredom together. -1 111 > In the end, that's not such a bad thing to share.