The state of AI

This is a fascinating chart of how quickly new technology can be adopted.

It’s all just wham, zoom, smack into the ceiling. Refrigerators get invented and within a few years everyone had to have one. Cell phones come along, and they quickly become indispensable. I felt that one, I remember telling my kids that no one needed a cell phone back in 2000, and here I am now, with a cell phone I’m required to have to access utilities at work…as well as enjoying them.

The point of this article, though, is that AI isn’t following that trajectory. AI is failing fast, instead.

The AI hype is collapsing faster than the bouncy house after a kid’s birthday. Nothing has turned out the way it was supposed to.

For a start, take a look at Microsoft—which made the biggest bet on AI. They were convinced that AI would enable the company’s Bing search engine to surpass Google.

They spent $10 billion dollars to make this happen.

And now we have numbers to measure the results. Guess what? Bing’s market share hasn’t grown at all. Bing’s share of search It’s still stuck at a lousy 3%.

In fact, it has dropped slightly since the beginning of the year.

What’s wrong? Everybody was supposed to prefer AI over conventional search. And it turns out that nobody cares.

OK, Bing. No one uses Bing, and showing that even fewer people use it now isn’t as impressive a demonstration of the failure of AI as you might think. I do agree, though, that I don’t care about plugging AI into a search engine; if you advertise that I ought to abandon duckduckgo because your search engine has an AI front end, you’re not going to persuade me. Of course, I’m the guy who was unswayed by the cell phone in 2000.

I do know, though, that most search engines are a mess right now, and AI isn’t going to fix their problems.

What makes this especially revealing is that Google search results are abysmal nowadays. They have filled them to the brim with garbage. If Google was ever vulnerable, it’s right now.

But AI hasn’t made a dent.

Of course, Google has tried to implement AI too. But the company’s Bard AI bot made embarrassing errors at its very first demo, and continues to do bizarre things—such as touting the benefits of genocide and slavery, or putting Hitler and Stalin on its list of greatest leaders.

Yeah. And where’s the wham, zoom, to the moon?

The same decline is happening at ChatGPT’s website. Site traffic is now declining. This is always a bad sign—but especially if your technology is touted as the biggest breakthrough of the century.

If AI really delivered the goods, visitors to ChatGPT should be doubling every few weeks.

In summary:

Here’s what we now know about AI:

  • Consumer demand is low, and already appears to be shrinking.
  • Skepticism and suspicion are pervasive among the public.
  • Even the companies using AI typically try to hide that fact—because they’re aware of the backlash.
  • The areas where AI has been implemented make clear how poorly it performs.
  • AI potentially creates a situation where millions of people can be fired and replaced with bots—so a few people at the top continue to promote it despite all these warning signs.
  • But even these true believers now face huge legal, regulatory, and attitudinal obstacles
  • In the meantime, cheaters and criminals are taking full advantage of AI as a tool of deception.

I found this chart interesting and relevant.

What I find amusing is that the first row, “mimics human cognition,” is what I’ve visualized as AI, and it consists entirely of fictional characters. They don’t exist. Everything that remains is trivial and uninteresting — “Tinder is AI”? OK, you can have it.


  1. KG says

    AlphaFold. Yeah, right, dramatic improvements in the prediction of protein folding from the amino acid sequence are completely trivial.

  2. cartomancer says

    People say that mobile phones are indispensible, but I disagree. I was given one a few years back, but I’ve never found it at all useful. Perhaps if either of my friends ever responded to the messages I send them it might be. These days it’s only vaguely useful function for me is as a travel alarm clock.

    But hey, procedurally-generated pictures of Victorian lawyers having a snowball fight! That’s fun, right?

  3. JM says

    The chart doesn’t look to be exactly fair. The base seems to be when the technology became really practical, not when it was first invented. People were working on TV since around 1900. Purely mechanical systems had too many problem though and didn’t go anywhere. All electronic systems came around 1930 but still had problems and patent log jams between the major companies got in the way. The 1950 date they list is roughly when the US got unified standards for broadcast. I can see the same sort of thing with several other items listed.

  4. christoph says

    Looks like we won’t have to worry about the imminent rise of the machines for a while.

  5. lotharloo says

    I’ve thought about the issue of overhyping AI and I think the best example that I can think of is this:
    Imagine a civilization where there are no cars and everyone is either walking, running, or riding horses and donkeys for the purposes of transportation. Now imagine that a research group invents cars. The invention revolutionizes a lot of the aspects of the civilization but very few people understand how cars actually work. Then you have influencers, politicians, philosophers who don’t really have any technical understanding start promoting cars as a “magical transportation device that given the speed of their development will eventually manage to take anything from any point A to any point B.” And then you have people trying to calculate the day everyone will be driving their cars to the moon. You have companies auctioning lands on every planet in the solar system and then fighting over the land rights on exoplanets, because soon enough we all will be driving our cars there.

    Right now, “AI” is in a similar situation. It’s a very useful tool but it’s overhyped as fuck. The decline in chatGPT traffic is extremely expected and normal. The product was incredibly overhyped, and often irresponsibly by openAI. People actually think AI is intelligent, i.e., human intelligent. Very few people know that the basic building blocks of all these “AI”s cannot solve trivial math questions such as counting whether there are odd or even number of things.

  6. ardipithecus says

    In my newsfeed a few days ago, there was a report of a judge ruling that ai generated content is not copyrightable, The legal war has begun.

  7. Walter Solomon says

    Everything invented before the 1950s seem to really take off in the post war prosperity following WWII. Everything invented after the 60s seem to get adopted comparatively instantaneously.

    The trajectory for video games is interesting. They start off very popular in the late 70s/early 80s but by the mid 80s the market has a glut of poorly made, crap games. This nearly destroys the video game industry and it was only rescued by Nintendo’s NES.

  8. whywhywhy says

    AI has made a huge impact in medical imaging and is now everywhere which was not the case a few years ago. For example, look at AI automated image segmentation. The need to manually segment images to permit further analysis kept many discoveries and inventions from being implemented in a clinical setting, thus reducing the accuracy of diagnoses. In most cases this segmentation can now be performed by computers in seconds rather than by humans over hours and days. This enables more and better information for patients and clinicians when they need it.

    Calling AI a complete failure based on whether it improves search is myopic. This is like saying saying TV has failed since nobody is placing screens in dairy barns to improve milk production. None of these technologies are good for every application.

  9. Walter Solomon says

    What I find amusing is that the first row, “mimics human cognition,” is what I’ve visualized as AI, and it consists entirely of fictional characters.

    Not to mention Terminator is a poor choice for that axis. It only mimicked/exceeded humans physically but not cognitively. It was indistinguishable from humans while walking around and was stronger but it was pretty easily outsmarted. The second film even poked fun at how bad it was at comprehending human emotions.

  10. billseymour says

    I will soon be starting radiation treatment for a very small bit of cancer in my lung.

    They gave me a CAT scan last Wednesday, I gather to make sure that they know exactly where the growth is, and to generate data that will be fed to some kind of software that will point the radiation at the right spot; and now I’m wondering whether there might be a bit of AI involved.  Since I’m a computer programmer by trade (now retired), this is interesting to me.

    I’m going in for a “dry run” this afternoon to make sure that everything is working as expected; then I’ll have five actual treatments on Friday, then Monday through Thursday.

  11. Dunc says

    Generative AI is not the whole of AI. Sure, it’s the bit that’s getting all the hype right now, but AI – or rather, machine learning – is much, much broader, and is doing all sorts of things very well indeed. But it’s not going to write you a poem or have a conversation.

  12. robro says

    One of these things is not like the other. You’re comparing the progress of developing things and “adopting” them…whatever that is…to something you are calling “AI”. Like many graphs there are some problems here.

    For example, the line for cars starts around 1915 and then hits 100% “adoption” in 35 to 40 years. The first recognized “practical” modern automobile was the Benz Patent-Motorwagen. It came out in 1885. So a more accurate time line is more like 65 years. But, perhaps it’s even more. The first steam automobile-like things were steam carriages produced in the 1700s. If we include those, then the time line from conception of a self-powered vehicle to the “adoption” of the automobile is more like 200 years. Along the way there were many fails. No one is driving a steam ICE these days that I’m aware of. The Edsel was a huge fail for one of the more successful producers of automobiles even after 1950.

    Also, many of the things on the right with their steep ramp ups are all built around the same thing: silicon chips…essentially different implementations and products based on that technology. By the way, these silicon-based electronic products are often designed and frequently produced using “computer-aided design” where various forms of “AI” are part of the tool kit.

    So it’s a skewed graph. What else is new about graphs?

    A key here is comparing the development of things, specifically consumer products, to a collection of algorithmic techniques. People may never go down to a store to buy some “AI” or hold some “AI” in their hand per se. If you give “AI” 200 years, our descendants might buy a protocol droid or a HAL but that’s crystal ball gazing. Yet, “AI” techniques are already being used in engineering, science, medicine, and other areas. The “AI” world is not defined by Microsoft Bing, chatGPT, Google, or Facebook.

    The biggest hurdle for forms of “AI” like chatGPT based on Large Language Models (LLMs) or any other technique of automated language analysis and classification is rooted in one word: Language. Language is slippery, sloppy, complex, diverse, and ever-evolving…which I’m sure you know. I happen to be working, sort of, in this area. Whether these experiments produce anything useful for the business I’m in is very much TBD. Whether it ever helps put something useful in front customers is even more TBD.

  13. chrislawson says

    AI as a field is not failing at all. It is in the middle of a technological boom that goes mostly unnoticed by the general population. What is failing is AI for marketing datamining and wage suppression pushed by unethical companies.

  14. Dunc says

    And, good lord, but that’s a bad chart! Why is it only air travel (at least, I think it’s air travel*) that starts at zero adoption? Why do all the others start out around 10%? Could it be that starting from zero would make it obvious that inital adoption of new technologies is often very slow?

    Claim from the OP:

    Refrigerators get invented and within a few years everyone had to have one.

    From wikipedia’s article on refrigeration:

    [I]n 1834, an American expatriate to Great Britain, Jacob Perkins, built the first working vapor-compression refrigeration system in the world. It was a closed-cycle that could operate continuously[…] His prototype system worked although it did not succeed commercially.[11]

    In 1842, a similar attempt was made by American physician, John Gorrie,[12] who built a working prototype, but it was a commercial failure.


    The first practical vapour-compression refrigeration system was built by James Harrison, a British journalist who had emigrated to Australia. His 1856 patent was for a vapour-compression system using ether, alcohol, or ammonia. He built a mechanical ice-making machine in 1851 on the banks of the Barwon River at Rocky Point in Geelong, Victoria, and his first commercial ice-making machine followed in 1854. Harrison also introduced commercial vapour-compression refrigeration to breweries and meat-packing houses, and by 1861, a dozen of his systems were in operation

    But wait! You meant the invention of the domestic refrigerator, right? And yes, they were adopted very quickly – but that just serves to illustrate the difference between the initial development of the technology itself, and the development of successful consumer products based on that techology – which in the case of refrigeration, is a difference of around 100 years.

    Can I just mention how much I hate charts with lots of squiggly lines in very similar colours?

  15. says

    I suspect refrigerators are a special case, because they’re so universally useful. No matter who you are or where you live, you need fresh food.
    And then there’s the fact that one technology relies on first adopting another. Note that the plateau in electricity adoption ends just around when refrigeration starts taking off. Probably not a coincidence.

  16. wzrd1 says

    I think that air travel was an excellent example, adoption was slow, due to both a high fatal crash rate and high cost for air travel. It wasn’t until the crash rate dropped quite a lot and costs descended from the stratosphere that air travel grew in acceptance for everyday usage.
    But, with AI, we had all manner of trumpeting how it’ll be everything for everyone, hell, I still get hucksters e-mailing me on how chatGBT can magically improve my resume and that’s right up there in my utility index with flying cars. They can’t drive worth a fuck on the ground, so I’d want them zooming about overtop of my head?!
    AI is highly limited in capability currently and that isn’t going to disappear, it’ll remain common. It’s phenomenally, even revolutionarily useful in certain applications, but it’s nothing like what the marketing crowd keeps incorrectly asserting (largely, I suspect, to attract investors to their vaporware). Machine learning, sorting complex datasets for specific events and trends, examining vast and complex datasets for changes, phenomenally useful. Facial recognition, conversation, getting the vast content of the web sorted correctly on a spur of the moment search with accurate results, not so much.

    As for search engines, I was happy with RegEx searches, I can get results essentially instantly for what I wanted, instead I now get bombarded with commercialized bullshit, with popular preferred as a result – after paid promotions, limiting the value of the leading search engines to pretty much on par with the value of sorting the bit bucket. Who’s recently searched /dev/null? Anyone?
    As an example of useful, Google did lead the way in natural language search, while retaining RegEx capabilities, basically converting natural language searches into RegEx behind the scenes, yielding effective results. Then, they started going more toward paid content leading, followed by popular content, which can trivially be gamed (and is done so by search optimization firms), rendering decreasing value from any results. When you have to page down more than once to get past the bullshit, your return on expenditure of effort has been minimized. Efficiency goes out the window, as I’m stuck then with sorting through advertisements and popular results, rather than what I was actually sorting for – which is precisely what the service was supposed to provide. It goes into noise to signal, with the results having far too much noise to really impart value.

  17. says

    PZ wrote: ‘I do know, though, that most search engines are a mess right now, and AI isn’t going to fix their problems.’ And, thankfully, he is aware of how terrible G00GLE is.
    I reply: I agree with that. Their results are often compromised by deceitful website metadata*. Artificial Intelligence has been proven to only muddy things more. G00GLE is spyware crap that just feeds you more of what it has tracked that you like. However, duckduckgo is at least not heavy spyware and can be tailored to suit your needs.
    We are not luddites, but we strictly limit the control software and AI have over our lives.
    *As an example, I was looking for vaccination data and in duckduckgo I stipulated ‘only from the past month’ However, many of the results were from 2 or more years ago. The sites carefully prevented accurate dating by the search engine. What Bullshit!

  18. Chakat Firepaw says

    @JM #5

    The chart doesn’t look to be exactly fair.

    It’s not: Except for air travel, it’s cutting off the bottom of the logistic curves where there is little growth while the technology is developed and use cases are explored. While some of those things had a very quick entry to the rapid adoption phase, some sat as a niche technology for decades.

    Generative AI is in that developmental stage right now: It’s a toy with a practical use for a handful of people, (e.g. one artist I follow uses an image generator trained on his art to handle what he considers the ‘just work’ part of drawing so he can spend more time on composition and detail work).

  19. says

    @14 robro said: No one is driving a steam ICE these days that I’m aware of.
    I reply: I know it’s pedantic, but steam vehicles were ECEs (external combustion), not ICE. And there have been experimental Stirling powered vehicles that were much safer and could run on any heat producing method. Many of my cohorts drive only Electric Vehicles (often home built). More to the point, adoption of technology by people has not always been ‘intelligent’. Sony Betamax and ElCassette were two glaring examples. There have always been stupid and sometimes dangerous applications of technology. The futile tug-of-war among M$ windows, Appple and Chrome point out the idiocy of the ‘market’. We use Linux in many forms and for almost every computing need. Pushing AI into everything is mostly absurd, but it is popular.

  20. Andrew Dalke says

    I agree with others here about the difficulties of placing AI on a chart which, for the most part, requires 10% “adoption” rate before making the chart.

    I tried to figure out what “adoption” meant. If I go on one flight in my life, flowed by a visiting barnstormer, is that adoption of air travel? If I take a commercial flight once in my life, is that adoption? If I tried ChatGPT once, is that adoption? Are the numbers per person or per household? Surely some, like microwave adoption, are per-household. (Otherwise, how can the credit card numbers be 90% when 18% of the population is under 15?)

    The source seems to be , which does not describe where the numbers came from.

    I would like to see bicycles on the same plot. After the safety bike was invented, there was a bike boom in the 1890s and early 1900s, which deeply transformed society.

    At least some 9 million bicycles manufactured during that period, if I interpret correctly. The US population in 1900 was 76 million people and 26 million were 14 or younger, so unlikely many got a bike costing ~$2,000 in today’s money. I see statements like 20% of Chicago rode a bike during that time.

    This shows bicycle adoption during the 1890s would surely make the chart, and likely have grown faster than any of the pre-WWII curves shown, and faster than several of the post-war ones.

    Yet, and this is important, there was a big slump in bicycle sales in the early 1900s. The old bike craze was over and the new craze was the car. According to the linked-to page, those curves show the “demand pattern for real innovation”, and not following those curves is an indication of a “fad or craze” – which, yes, is what bicycles were, but the fad was on top of a deeper demand.

    The adoption rate for bicycles is now quite high, yes? Even though there’s only been gradual innovation in bike technology over the last century.

    I wonder how many other counter-examples there are to this curve model. At the very least, this bicycle example shows the difficulty of using short-term changes in adoption analysis.

  21. John Morales says

    In the news:

    Technology giant Nvidia says its sales have hit a record after more than doubling as demand for its artificial intelligence (AI) chips soars.

    The company says revenue jumped to above $13.5bn (£10.6bn) for the three months to the end of June.

    Nvidia also expects sales to soar further in the current quarter and plans to buy back $25bn of its stock.

    The firm’s shares rose by more than 6.5% in extended trading in New York, adding to their huge gains this year.

    Nvidia also said it expects revenue of around $16bn for the three months to the end of September.

    That is much higher than Wall Street expectations and would equate to a rise of around 170%, compared to the same time last year.

    “A new computing era has begun,” Nvidia’s chief executive, Jensen Huang, said in a statement.

    “Companies worldwide are transitioning from general-purpose to accelerated computing and generative AI,” he added.

  22. anat says

    cartomancer @2: I could probably replace all the functions for which I use my phone with a combination of a lap-top computer and a flip phone, but then I would have to lug a lap-top around everywhere. The phone weighs less.

  23. Rob Grigjanis says

    Is Your Smartphone Making You Less Smart? Distraction Addiction Is Real

    In a 2014 study published in the journal Science, researchers put their subjects in an empty room for 15 minutes, with no access to reading material or technology, offering the subjects the option of administering themselves a small electric shock. Two-thirds of the men and one-quarter of the women actually shocked themselves, preferring that jolt of electricity to being alone with their thoughts.

    It’s probably my COMA (Cranky Old Man Affliction), but I also find people continually using their cell phones annoying as hell.

  24. says

    re: BillSeymour #12

    I hope your treatments help. It sounds like you are going through a nightmare scenario but that you are doing your best to cope with a difficult situation.

  25. Rich Woods says

    @Rob #26:

    Two-thirds of the men and one-quarter of the women actually shocked themselves

    I bet half the men did that entirely because they wanted to see what it felt like.

  26. says

    Well before cell phones I made sure to have a book with me wherever I went. I even kept a couple in the car for emergencies.

    Now the majority of my library is on my phone: all I have to carry is a cell phone and a small power bank for longer trips.

  27. says

    @32 mikeschmitz said: duckduckgo IS bing
    I reply: your simplistic statement smacks of confusing the map with the terrain and is a big mistake. First error, Duckduckgo is not just a search engine, it is now a browser and you have not differentiated them. Second, almost every search engine relies on one of about 3 major databases. Just because Duckduckgo draws raw info from the bing database does not mean they are the same. Duckduckgo does a lot of filtering and purification as in the following report:
    Search privacy company still needs Bing, but won’t allow Microsoft’s trackers.
    Kevin Purdy – 8/5/2022, 2:33 PM
    “Previously, we were limited in how we could apply our 3rd-Party Tracker Loading Protection on Microsoft tracking scripts due to a policy requirement related to our use of Bing as a source for our private search results,” Weinberg writes. “We’re glad this is no longer the case. We have not had, and do not have, any similar limitation with any other company.”

  28. wzrd1 says

    @ 21, I’m aware of two steam ICE engines. The V2 rocket had water diluted ethanol for fuel, the water cooled the combustion chamber and provided additional reaction mass. Water injection on passenger jet engines, to again, provide additional reaction mass for takeoff.
    Seriously niche uses, to put it mildly.

    @ 25, my smartphone, with a docking adapter/station can hook to a monitor, mouse and keyboard and be used like a weakish desktop. Never really caught on. One could also run one’s phone apps via software linking both through the USB-C, although Samsung dropped Linux support quite quickly, screwing themselves out of one significant market.
    My next smartphone very well may end up being a Linux based phone, just so that I still have a quick and dirty desktop in a pinch when traveling.

  29. says

    @34 wzrd1 said: ’m aware of two steam ICE engines.
    I reply: yes, you are, of course. correct in their use of water. But, I was only considering the context of the earthbound vehicles that robro was discussing.
    @34 wzrd1 said: My next smartphone very well may end up being a Linux based phone,
    I reply: I have been following the development of those and really like the idea. They are not quite to the point of being fully useful at a price I can afford. I have found a bunch of good reviews of that hardware and software on Distrowatch, too.

  30. Chris Whitehouse says

    I’m using AI in Photoshop dozens of times every day. It is giving me a powerful tool that I couldn’t have conceived of a year ago, greatly enhancing my work in photo restoration.