George Carlin VS state-of-the-art AI

My current take on recent AI development is that it is getting more and more useful.

BUT, in the end, it is still a glorified, but novel statistics [1].

And whenever AI learns from human input, I find funny/insightful/dark quote from George Carlin:

Or in other words, current crop of AI cannot escape law of large numbers [3].

Especially in today’s world, you need very-very-very valid input sources vetted, by of course, error prone humans.
Even reviews for the simplest of stuff you can buy cannot be trusted as they are bought in bulk.
And as misinformation efforts are running loose in the wild, it is hard to keep sources clean.

Even for pure technical domains such as programming, ChatGPT has been banned by StackOverflow [2] due to high percentage of only looking correct answers.

And some fun chat-bot responses from history characters 🙂

https://twitter.com/RealESonneborn/status/1615794316504440839

Conclusion

So until AI can learn different real-life models instead of shoveling data into hundreds of billions of (statistical, black-box) parameters with insane compute power needs, it will just be useful statistics.

But there is a lot of room for research in how even today’s AI works, even on much smaller scale models.

References

[1] – https://towardsdatascience.com/no-machine-learning-is-not-just-glorified-statistics-26d3952234e3

[2] – https://meta.stackoverflow.com/questions/421831/temporary-policy-chatgpt-is-banned

[3] – https://www.investopedia.com/terms/l/lawoflargenumbers.asp

Weekly breakdown – 22w20d2

Security

https://research.nccgroup.com/2022/05/15/technical-advisory-tesla-ble-phone-as-a-key-passive-entry-vulnerable-to-relay-attacks
https://duo.com/decipher/researchers-demo-relay-attack-against-bluetooth-le-systems
More and more successful real-life attacks will happen on Bluetooth protocol features unless additional layer of security is used.

Even though security was thought of when Bluetooth was originally introduced (year 1998.), state of security considerations and possible attacks were not the ‘most important’ items on the list. Further improvements resolved some problems, but fundamental issues with the protocol itself cannot be solved without breaking backward compatibility.

As one of the speakers at EU Tech Chamber session about IoT vulnerabilities last year, I had luck to listen to presentation from Maximilian about broken fundamentals in Bluetooth standard.
Just check KNOB Attack how easy is to manipulate Bluetooth key negotiation to reduce attack space.

Artificial intelligence

https://www.nature.com/articles/d41586-022-00858-1
AI is becoming more and more important tool in all kinds of applications.

For any application that touches safety, security and actual real-life implications, it should be paramount to understand why AI system came to the specific conclusion, something like a train-of-thought.
.
Imagine (near) future situation when (it is not if) autonomously driven car causes an accident, it would be beneficial to have such kind of logs when investigating without complex proprietary technology needed for each car vendor.
Or imagine medical AI assistant decides that some person has some illness, the doctor should have information why it has deduced such conclusion. Sometimes AI assistant would be wrong, but sometimes it can actually remind doctor about different possibilities.

I would not be surprised, and probably it should be pushed for quite heavily, that some kind of legal framework is put on top of safety and security related AI systems in which they are required to produce something like standardized decision-tree for their actions or inactions.

Random

https://www.newscientist.com/article/2319584-computer-powered-by-colony-of-blue-green-algae-has-run-for-six-months/
With current prices of electricity, it is not not so far fetch idea to have your own pond of algae on stand-by, just in case 🙂