Beep boop - this is a robot. A new show has been posted to TWiT…
What are your thoughts about today’s show? We’d love to hear from you!
Beep boop - this is a robot. A new show has been posted to TWiT…
What are your thoughts about today’s show? We’d love to hear from you!
Still listening to the episode, but I just have to say: Clue is one of the greatest movies of all time. It’s very silly and amazingly quotable.
Despite a few faults that being a movie from the 80’s can bring, I think it still holds up today!
Holy cow - these ladies rock! This is the kind of meat I look for on my tech podcast. Outstanding!
Great show this week, although Leo kept trying to butt in at certain points. I’m glad they had enough confidence to finish what they were saying, many guests don’t. (I know this is probably mainly due to lag, I’m not getting at Leo)
Just going to be honest, even Bard is too good a writer to create a Seinfeld episode, that was probably the most overrated show ever. - it was the James Corden of sitcoms!!!
Unfortunately Amanda had significant (like several seconds) lag. But we managed and I agree the panel was great!
Cathy speaking about automated transcribing jogged a memory for me.
Once upon a time, I worked as a cub reporter at a podunk radio station. Most of the time, my assignment was the crime blotter, but I got occasional assignments for local government. One of the early ones was a mayoral debate.
I can only imagine how much help it would have been to have automated transcription and synopsizing of the debate. Or for that matter, any public meeting. Both for accuracy (getting quotes exactly right) and speed.
I’m bemused at how many people are latching onto a spiritual argument against the possibility of AI art being meaningful and/or sapient AI ever existing in the future. Even total atheists are falling back on arguments that are just a couple of synonyms away from saying “humans have souls and machines, of course, do not.” Call it je ne sais quoi or whatever else, but I have yet to see any difference between this popular belief and our long, ongoing struggle to classify animal intelligence levels, and how efforts within that science have been biased by both anthropomorphism and cultural spirituality.
And I continue to be very concerned about how our determination to draw lines between artificial intelligence and human intelligence is already leading to those lines being drawn where they effectively classify some neurodivergent people as non-human.
I think it’s even simpler than that. I could use a recording device to record you “performing” your signature. Then I make a robot that performs it on command (say you’re famous and are sending out autographed photos). You as a human are presumably aware of what you’re doing, and potentially proud of the fact that people are requesting your autograph. I don’t know when we’d ever get to a sentient AI that is self-aware enough to have the same sense of pride in an action so simple as making a signature. Also, you’re not a precise enough machine that every time is exactly the same, whereas it’s probably going to require a human design injection into a robot to allow it to “vary” the signature slightly every time, because otherwise it will more or less just be a photo copier.
Is emotion required for sapience/sentience?
Well emotion is [generally] an external display of state. It’s one of the few ways we have as fellow humanoids to get a sense of what the brain of the other humans might be experiencing. The real point is that without self-awareness, where you ACTUALLY understand what you’re doing, not just pretending to be a robot following instructions, no matter how complex, I don’t think you have sentience.
A sense of purpose, more than emtions.
At the moment LLMs (which aren’t AI), for example try to answer questions by stringing together words based on probability. They don’t know what they are talking about, they don’t know whether what they are saying makes any sense, they don’t know whether what they are saying is fact or fiction. They just “roll the dice” and come up with an answer they believe makes sense, based on what they have learnt from the LLM they were trained on.
This is not intelligence, let alone sentience. I don’t want to go all Butler on us an start a Jihad to wipe out the “thinking machines”, before they even exist, let alone managed to gain a foot hold in our society, but some sort of balances and measures needs to be put in place, maybe not quite a Turing Police, but certainly an independent ethics commission that overseas LLMs, GANs, DL and whatever comes up next.
My problem isn’t with the technology itself, but its marketing and implementation in the public space. Some things, like some of the co-pilots that Microsoft has shown/proposed, seem to be sensible. DL and GANs in photography and videography to help smooth out images or remove unwanted artefacts, for example, a method of asking for data to be analysed based on plain English that then formats it in Excel, with the relevant formulas, fine.
Where I have my problems is with ChatGPT, Bing AI, Bard etc. these things are still in their early stages and are appropriate for the lab for study, but they aren’t ready to be released on a gullible public as “speakers of truth”. Their wording has been trained to be convincing, whether the answers are correct or not. Microsoft is aiming Bing AI as an alternative to search results, but its biggest problem is, it provides made up facts and delivers them with authority.
My first ever attempt at a Bing AI search provided financial information on cash transactions, but its answer was out by a factor of 10! Following Bing AI’s advice would get somebody probed by the Finanzamt (IRS) for breaking the law.
Another time, I asked it for the lyrics to a children’s song sung in our region. Instead of saying it didn’t know, it made up the lyrics (or more likely, it found a video of children going door to door & singing out of harmony and transcribed all the voices in parallel). Luckily I knew enough of the lyrics to spot immediately that the answer from Bing was completely wrong. In the end, I managed to find it using DuckDuckGo and I could train Bing with the correct answer. But if someone who had never heard the song asked, they might think the words looked a bit funny, but, being in Platt and not High German, they might just assume it looks strange, because it is in dialect, not because it is totally wrong!
Here is what it suggested:
Mattemännken goue man Goue man, goue man Mattemännken goue man Goue man, goue man Mattemännken goue man Goue man, goue man
Wat häst du in de Tasse? In de Tasse, in de Tasse Wat häst du in de Tasse? In de Tasse, in de Tasse Wat häst du in de Tasse? In de Tasse, in de Tasse
Kaffee un Kauken Kauken un Kaffee Kaffee un Kauken Kauken un Kaffee Kaffee un Kauken Kauken un Kaffee
Dat will wi nich hebben Nich hebben, nich haben Dat will wi nich haben Nich haben, nich haben Dat will wi nich haben Nich haben, nich haben
Wat willt du denn hebben? Denn haben, denn haben Wat willt du denn haben? Denn haben, denn haben Wat willt du denn haben? Denn haben, denn haben
Äppel un Birnen Birnen un Äppel Äppel un Birnen Birnen un Äppel Äppel un Birnen Birnen un Äppel
Dat kanns du wol kriegen Wol kriegen, wol kriegen Dat kanns du wol kriegen Wol kriegen, wol kriegen Dat kanns du wol kriegen Wol kriegen, wol kriegen
Nu giff mi dat Äppelken Dat Äppelken, dat Äppelken Nu giff mi dat Äppelken Dat Äppelken, dat Äppelken Nu giff mi dat Äppelken Dat Äppelken, dat Äppelken
Nu giff mi dat Birnchen ook Dat Birnchen ook, dat Birnchen ook Nu giff mi dat Birnchen ook Dat Birnchen ook, dat Birnchen ook Nu giff mi dat Birnchen ook Dat Birnchen ook, dat Birnchen ook
Nu dank wi di so schön dafür So schön dafür, so schön dafür Nu dank wi di so schön dafür So schön dafür, so schön dafür Nu dank wi di so schön dafür So schön dafür, so schön dafür
Un wünsch di allns wat goot is Wat goot is, wat goot is Un wünsch di allns wat goot is Wat goot is, wat goot is Un wünsch di allns wat goot is Wat goot is, wat goot is
Un wenn wi wedder kommen Wedder kommen, wedder kommen Un wenn wi wedder kommen Wedder kommen, wedder kommen Un wenn wi wedder kommen Wedder kommen, wedder kommen
Dann willt wi mehr hebben Mehr hebben, mehr hebben Dann willt wi mehr hebben Mehr haben, mehr haben Dann willt wi mehr haben Mehr haben, mehr haben
The actual lyrics are as follows:
Mattemännken gohe Man,
ob du uns was geben kann,
Appel odder Bireren,
Nörte sind vögierwen,
aule Statd - nigge Stadt,
leewe Jungfrau giff us wat,
laut us nich tou lange staun,
wi mört noch’n Hüsken wiedergaun,
bet nau Briärmen,
do wült wi us’n bittken wiärmen,
bet nau de nigge Stadt,
do stiärk’t se us oll tohaup innen Sack
That is the problem with so-called AI that we have at the moment. It is fine for use in the lab, but it isn’t intelligence and we have a very long way to go, until it becomes sentient. We probably have a couple of decades, at least, before we need an answer to that question - which gives us plenty of time to think about it and come up with a good definition.
This is probably the 5th big “AI wave” I’ve experienced, since I got into computing, with “real AI” coming soon. The hype and the scale is bigger this time, but I still don’t see any real AI going on out there.
So does Google, plain old boring Google, it will give you whatever cr@p has been paid or played into it’s algorithms top 10 spots. The gamification of Google has made it no better that any alternative engine, heck it used to more satisfying using the Yahoo directory.
There is a chance with AI that it could be used to build a secondary index based upon aggregation of primary search results initially it would take time, but eventually will produce significantly more accurate results than the current ChatGPT and classic search engine models
I do find it amusing when people say that things aren’t AI. With the utmost respect and politeness, I would proffer that they aren’t your personal definition of AI.
Many years ago I designed an built a system called ‘ADoLF’ Automated Diagnostic of Logic Faults.
I was designed for testing channel cards in a digital mixer. With 108 cards in a system it would take weeks to fully test.
The role of these cards was to carry out an FFT on an incoming digital signal enable volume and frequency response of the digital path to be monitored. Using a digital pattern generator as an input it was possible to predict the absolute output, it was also possible using specific pattern pairs to identify problem areas and components on the card. Full manual testing took 3 days per card, fault finding about twice that. ADoLF took that down to under 15 minutes, half of which was loading the card and the software.
It took my design and engineering intelligence focused it on task and saved days IMHO that was Artificial Intelligence
What you described was an automated test harness, the description doesn’t sound like AI.
Could the test harness look at the card, work out what were valid and invalid inputs and outputs? Could it learn when new, never seen before, errors cropped up and deal with them accordingly? That would be machine learning, but not really AI.
AI is bandied about with gay abandon at the moment, but most of the systems have little or nothing to do with “real AI”.
I don’t expect Google (the search engine) to do anything more than give me links to evaluate. I don’t consider it to be a beacon of truth, but a path finder to something that I can evaluate for truthfulness. The problem I have with people proposing the ML tools of today as oracles, is they’re not vetted, and I don’t know how one could vet them fully, since part of their operation is little hits of randomness. Google, at least, seems to give the same results for the same inputs, in general, with searches within a certain timeframe (given they’re probably constantly indexing new information.) And lets not get started on whether or not we should trust Wikipedia… I spent a good couple of hours defending it from vandals trying to make a living certain person dead for whatever perverse reason.
It was able to parse the outputs and then alter the patters based on the outputs to guide testers to parts of the circuit with errors and discover dead components. You could not give the test and fault finding process to an entry level engineer, but with ADoLF they could.
Then you have the previous product I worked on the DSI9000 Tempest breach system which was able to monitor RFI noise and identify systems based on pulse signatures and then present a range of options to breach the system, comprehensively more powerful than most human engineers, leveraging the speed of the computer analysis
However I would say guys are the exception, many people including a number journalists and podcaster (not @Leo) believe Google to be the gospel truth, you only have to listen to the gang at the Vergecast