Back in September 2023 I noted that the biggest hurdle for AI completely replacing physicians is the physicality of the job. Sure, LLMs are good at giving differential diagnoses and faking empathy once somebody’s problem has been reduced to text, but the art of medicine is in the act of seeing, feeling, smelling, etc. [Note: Although increasingly less so, as doctors and trainees are becoming experts at treating patients in the chart and not those in front of them, making themselves the perfect foils for replacement; queue photo of the old man yelling at clouds. ] If clankers have any hope of replacing humans, they’d better get some senses.

At first glance, a recent Nature Medicine paper aimed to do just that by introducing what the group of authors — all of them Google employees based in the UK and California — call “multimodal reasoning” but is in fact the chatbot being able to interpret images, ECGs and lab reports in addition to the pre-digested clinical pearl. The topline result, one that the journal itself felt obligated to headline, was that “AI had superior performance compared with physicians for almost every metric (29 of 32 axes)”. But at what?

You would think that the question would have been easy to answer, this being a peer-reviewed paper and all, but no. In fact, I am still not completely certain what interactions were performed and whether they completely match what was reported. What is certain is that a set of primary care physicians and patient-actors from Canada and India — countries different from the author’s own countries and let’s wonder conspiratorially for why that may be the case — interacted via an instant messaging-like service. This is the first oddity: even remote health visits are performed using video calls, and yes you may occasionally get a text through the EMR or if you are a VIP/boutique physician maybe your phone, but that is far from the norm.

The primary report is on what happened when the patients uploaded the skin photos, ECGs, lab results, etc. and then asked the physician or LLM on the other end questions about it. Pretty standard fare for a human-to-LLM interaction, but not exactly natural for a doctor-patient relationship which usually starts with questions being asked of the patient. This is the second way in which the setup was made to fit the computer and not the human.

But then the last section of the paper is about what happens when there is, in fact, a back-and-forth by the way of taking a history. The extended figures — “extended” here meaning not worthy enough of being included in the main paper — say it improves the performance of the LLM. They do not say how it affected the human performance, or how the patient-actors rated humans versus LLMs in history-taking. I would call that strike three.

To the journal’s credit, they did not allow Google to get away with it completely. “To evaluate the performance of our finalized system, we conducted a randomized, blinded human evaluation that emulates an objective structured clinical examination”, says the final paragraph of the introduction, only to end with:

We note, however, that our study is not a randomized clinical trial with prespecified endpoints and preregistered statistical analysis. Rather, it is an exploratory study investigating the properties of multimodal diagnostic dialogue.

Peer review is at least good for something, even if it does result in self-contradiction.

Meanwhile, in the world without motivating reasoning, more objective assessments of the usefulness of AI in medicine show that it is in fact still quite bad. This does not prevent the massively funded hordes of AI researchers from flooding the field with sloppy work, creating the impression that the rise of the machines is imminent. Comply or relegate yourself to the permanent underclass, serf MD. But of course, relegation will only be possible to the extent doctors — or any other profession, really — has already debased itself and abandoned its core professional principles in the service of electronic ease.

After dissecting the minutiae from the ongoing battle of the bozos ⊕ [Note: To save you a click: it is about the Musk-Altman trial. ] , Andrew Sharp’s weekly column ends with this paragraph:

The reality is knottier. Had the OpenAI founders not launched with a nonprofit structure in 2015, they probably never recruit the talent required to compete with Google. And had they done anything else other than exactly what they did in 2018 and 2019, all of computing would be less interesting today, and the company probably wouldn’t exist eight years later. Musk’s trial has been clarifying on that point, at least for me.

The AI side of technology is one of those rare occasions where biotech may indeed be like tech: people with knowledge, skills and ambition to make the early steps towards creating something new generally don’t do it for the money. Accolades, titles, a few more increments on their h-indices sure, but unless they are seriously delusional a lab postdoc coming in on a weekend to split the cell culture generally has no hope of getting into the top percentile in income. Up until a few years ago AI research was much like that, until it wasn’t.

Sharp writes that OpenAI had to flip the switch if it were to survive in these shark Google-infested waters once they smelled blood profit an opportunity to tell a new story to investors. Same can be said about any biotech: become successful enough, and there will come a time when the academic founders are asked to step away and let someone with different motivations run the show, lest they be lost in a sea of copycats, smoke-peddlers and competitive intelligence officers. The whole business has just become too expensive for some Jonas Salk-wannabe to dabble in.

A person of bad intent may propose that the adults coming to run the show once it becomes too expensive are the ones making it expensive in the first place to justify their existence, contributing the health care cost ouroboros on the way. But that is of course nonsense. The proof is in the pudding, what with famously efficient drug development pipelines, low health care costs and improving lifespans.

So let’s do what a genuine financial scion once proposed: invert. Instead of asking ourselves how to make drug development more efficient and cost-effective, let’s see how we could make it more expensive. Number one thing to do would making it all about the money: let’s portray people who don’t capitalize on their inventions as losers not heroes, make Nobel Prize winners notable only if they are billionaires (who won the Nobel Prize in Physiology or Medicine last year, again?), measure success of drugs in dollars earned not lives improved, extended or saved, have everyone skim a percent or five of the money swishing around in the ecosystem as their primary source of income without any penalty for ultimate failure ⊕ [Note: For more on this, do read Nassim Taleb’s Skin in the Game, which is about much more than the titular phrase which has become — much like his The Black Swan — a phrase people throw around without having any idea of the underlying concepts. ] guaranteeing that they will have every incentive possible to grow the pie, and I think you see where this is going because the system functions as designed so why should you complain? After all, there is no alternative.

Except that, of course, there is. It would be a big lift, to remove incentives of skimmers to inflate the balloon, stop various influencer platforms from inducing FOMO in everyone and anyone, recalibrate the median science journalist’s value system from Mr. Market to something more reality-based. Big, but not impossible, provided there is a will.

Therein lies the problem: that kind of thinking is somewhat at odds with the shared American culture, at least as recently described by Chris Arnade, that “you can live how you want, eat what you want, live (up to a point) how you want at a thin level, as long as you ultimately believe in making big money through hard work and playing by the rules.” Determining if the other two legs of the three-legged money/work/rules American stool are performing as intended I will leave as an exercise for the reader.

Though I agree with his stance on smartphones in schools and social networks in general, I fear that Jonathan Haidt of The Anxious Generation is now the useful fool for every US state senator with autocratic tendencies; Rindala Alajaji of the Electronic Frontier Foundation has a good write-up of how that came to be. Has parental guidance ever been successfully legislated?

• Ed Zitron: Am I Meant To Be Impressed?. A detailed account of why you cannot trust anything tech companies say about the business side of AI, even in official SEC-sanctioned documents (because, of course you can’t). Note in particular how the unsuspecting public buys their narrative about big AI spending leading to increased profits hook, line, and sinker (the “profits” coming from bloated valuations of OpenAI and Anthropic). I am no fan of Apple, but it appears that they have made the right choice by staying out.
• Richard Dawkins: Is AI the next phase of evolution? ⊕ [Note: ᔥJason Kottke ] A eulogy to New Atheism.
• Lada Nuzhna: What the hell is happening in China? This is with regards to their push in biotech. All good things, I’d say.
• Isaac Greene: The Gift of Music. A delightful reframing of gift-giving.
• Derek Sivers: Geography is four-dimensional. About knowing places only at a certain time. The short essay matches my own experience of being asked what Serbia is like now. Having last truly lived there more than 15 years ago, the only correct answer is that I haven’t a clue.
• Charlie Chaplin: The Final Speech from The Great Dictator, with transcript. It gave me chills.

To those who can hear me, I say - do not despair. The misery that is now upon us is but the passing of greed - the bitterness of men who fear the way of human progress. The hate of men will pass, and dictators die, and the power they took from the people will return to the people. And so long as men die, liberty will never perish…

Soldiers! don’t give yourselves to brutes - men who despise you - enslave you - who regiment your lives - tell you what to do - what to think and what to feel! Who drill you - diet you - treat you like cattle, use you as cannon fodder. Don’t give yourselves to these unnatural men - machine men with machine minds and machine hearts! You are not machines! You are not cattle! You are men! You have the love of humanity in your hearts! You don’t hate! Only the unloved hate - the unloved and the unnatural! Soldiers! Don’t fight for slavery! Fight for liberty!

Amen.

In the most recent episode of The Talk Show, John Gruber and MG Siegler agreed that the smartphone will be difficult to overthrow as the dominant method of computing. Something unthinkable would need to happen for us to leave the phones at home in favor of watches, earbuds or pendants, Her-style. So, even if SoC and batteries improve to such extent that we could fit the 2040s equivalent of an M5 chip into a MacBook, iPhone and AirPods equally, and all with great battery life, people would still reach out for their phones first.

This wasn’t the first time I heard the thesis, and it always sounded about right. I don’t know about everyone else, but I tend to be impatient when chatting with Siri. This isn’t about its “lack of” intelligence: although I had only used ChatGPT’s voice chats as a novelty when demoing it to elderly family members, even they needed a few more seconds to answer specific questions than my patience would allow. So why is that?

Well, my impatience would suggest that bandwidth is key, more specifically our own bandwidth to process information. Humans are visual creatures: much of our own brain’s neural pathways are tied up in receiving and processing information from the 6 million cones and 120 million rods contained in the approximately 2,200 square millimeters of our retinas. The next sense down in the number of receptors is not even close: touch, with about 4 million somatosensory receptors packed in the average 1.79 square meters — or 1.79 million square millimeters — of body surface area. That is two orders of magnitude more sensors packed into three orders of magnitude less space in retina (sight) versus skin (touch). What about sound, which is competing with sight as the interface of record? There are around 15,500 hair cells in each cochlea for 31,000 total — not even close. ⊕ [Note: This is why writing has been the defining achievement of our species, condensing the ineffable into something we can quickly process, and why I will never get tired of reading histories of notebooks and paper. ] But you don’t need to know any of this to have experienced dreams. When was the last time you remembered a particularly nightmarish smell or sound when you were asleep?

So if the optimal way for computers to communicate with us is via retinal inputs, does it not make the most sense to attach most of the local computing machinery onto the interface?

But what about our outputs? Unless you were born on Krypton one wouldn’t expect anything shooting out of your retinas to interact with the environment. Well, here is my main uncertainty in the smartphone-as-the-ultimate-device hypothesis: could you not, on an infinite timescale, wear contact lenses that could beam in information to you as efficiently as possible? Are the Apple Vision Pro and whatever creepiness Meta is out with now not steps towards our corneal computing future? Perhaps, but perhaps not, and the interaction with whatever is beamed into our eyes will be the next limiting factor.

All our movements are planned in the prefrontal cortex and executed in the precentral gyrus of the frontal cortex. A lot of that surface area is dedicated to our hands, as the creepy but to the best of our knowledge accurate cortical motor homunculus shows. ⊕ [Note: Fun fact about the brain: it’s plastic. In that, it can and does get rearranged as circumstances warrant, and the extent of the rearrangement can be drastic. But receptor numbers are what they are, so any broad changes to the general population would have to take… millennia? Dozens of millennia? Certainly longer than the life span — not to mention attention span — of the average S&P 500 company. ] The side by side representations of the sensory and motor homunculi twins is particularly striking in showing how important our hands are to our sense of self. Now, another prominent feature they have are large tongues and lips, much of it in the service of producing sound, so it is not a surprise that voice controls exist, and not just for the times when our hands are otherwise occupied. Yet what the homunculi show and what any cat parent will attest is that humans are, to the outside world, mostly a pair of hands attached to some rather bizarre squishy elements. Hey, we may as well own it!

The most serious consequence of this state of affairs, much to my disappointment, is that smartphones as we know them are here to stay so there isn’t much point in hobbling my own computing experience with black and white displays, “feature” phones and the like. While I would have loved, in my project to ditch Apple, not to replace my iPhone with anything else “smart”, reason says to try and find a true alternative. GrapheneOS looks promising, and by the time my current phone kicks the can or passes on something else may come along. Here’s hoping.

As promised, today’s update to Microbe — a micro.blog client for Emacs now at version 2.0 — includes draft syncing. There were also some minor updates to Inkling. Both are available on GitHub though I think I’ll just drop that and just host them here. Something to think about for next week…

Well, folks, I did it: I have hit FeedLand’s feed creation limit. Any chance this can be increased, @dave? I’ve just discovered a new blog I’d like to add.

I tend to avoid podcasts in the style of Joe Rogan, those that begin with a 15-minute long ad block selling mushroom supplements followed by hours of meandering conversation between two people who may or may not be under the influence. Who in the world has the time?

So for that reason I avoided the podcast of one Dwarkesh Patel even as I occasionally linked to an article of his. I filed him mentally in the same “Avoid!” bucket as Lex Fridman — probably unfairly, as no one in the world can be as big of a mental bore as Fridman — without giving his podcast a chance. Although, judging by his writing on AI, I would not have liked the tone even if I had heard it. I remember, in fact, resisting the temptation to pan some of his more outlandish texts prophesying the rise of our LLM overlords with a tone which was as matter-of-fact as it was uncaring about human culture and society. My headphones are a direct link to my brain and I did not want that kind of world view to influence it.

Well a whole bunch of people are about to get influenc’d, because the New York Times has just published a glowing profile of Patel and his podcast, framing the show as a way to “eavesdrop on the A.I. elite” while burying an important fact — the one that kept me from listening in the first place — in the fourth-to-last paragraph:

Mr. Patel doesn’t see himself as a journalist, and he will do things that news organizations’ ethics rules generally prohibit, such as signing onto an amicus brief on behalf of Anthropic in its recent lawsuit against the Department of Defense, and angel-investing in companies whose founders he has interviewed (he disclosed the stakes). He believes in a “glorious transhumanist future,” and his tone isn’t adversarial. But his admirers say that his technical fluency and extensive preparation enable him to follow up or push back on superficial answers that most interviewers would simply accept. The Jensen Huang episode became heated as Mr. Patel repeatedly challenged the world’s most valuable company’s chief executive on the national-security implications of selling chips to China. “If I do cover a topic,” Mr. Patel says. “I think my reputation would suffer a lot if I don’t ask tough questions or don’t do it in a deep way.”

Of course, praising for this kind of pushback on a transhumanist podcast is like praising the host of “The Ultimate Potato Chip Podcast” for pushing back against Frito-Lay’s most recent price hike: it goes without saying that you like junk food.

But it was not this small bit of confirmation bias which made me link to the NYT. Rather, it was the same revelation that piqued Tyler Cowen’s interest, if for a different reason. Rather than paste the whole excerpt, let me provide a (human) summary: bored during the covid pandemic, a 19-year-old Patel asks the libertarian George Mason economist Bryan Caplan to be a guest on his brand-new podcast; Caplan agrees. They continue the exchange, online and in person, while Caplan is spending months in Austin, TX at the home of his billionaire friend Steve Kuhn. ⊕ [Note: This wasn’t the only good billionaire-themed article in the NYT. For more reasons why Americans should probably do a bit more to clip their wings see the travails of one Sergey Brin and the series of hardships he endured that pushed him to the right. ] Kuhn also meets Patel and, liking the cut of his jib, offers to invest in return for equity. So do other people in the Caplan-Kuhn circle which inevitably expands all the way to your friendly neighborhood founder of Amazon. Cue NYT’s signature glazing.

Crikey. Fans of C.S. Lewis should recognize immediately the themes he raised in The Inner Ring, The Abolition of Man and That Hideous Strength, essays and books which were most likely not on Patel’s reading list during his formative years. One can only wonder whether his belief in “the glorious transhumanist future” came before or after the Silicon Valley billionaires made landfall in his young mind.

• Cory Doctorow: The enshittification multiverse, in which Doctorow proposes a general theory of enshittification to match his initial, special theory. I enthusiastically concur.
• Anonymous on the Marginal Revolution comments section: On health care price transparency. The only non-Xified content you can find on Marginal Revolution these days is in the comments, so I am glad that Cowen highlighted this minute dissection of the madness called American medical billing. Needless to say, I concur.
• Reese Richardson: A do-or-die moment for the scientific enterprise. ⊕ [Note: ᔥAndrew Gelman, who sure loves his mile-long headlines. ] This is the author’s summary of a more detailed paper in the academic journal PNAS which points to a looming catastrophe of LLM-boosted scientific paper mills holding hands with pliant journal editors to decimate the signal-to-noise ratio of the literature. Of course I concur!
• Cory Doctorow, again: Ada Palmer’s “Inventing the Renaissance”. His review after actually reading the whole book, and yep.

All I can think of while reading Nilay Patel’s software brain essay, quoted and linked to all over the web, is the slight but dense Metaphors We Live By. Software databases — metaphoric file cabinets and manila folders — now themselves becoming metaphors for physical objects is truly Escherian.