Impromptu — A Reluctant, Extempore Review
I read Reid Hoffman’s book, Impromptu, so you don’t have to.
Impromptu: adverb “Without preparation or premeditation; off-hand, on the spur of the moment; extempore.” noun: “Something composed or uttered without preparation or premeditation; an extemporaneous composition or performance; an improvisation.” from the Oxford English Dictionary
Huh.
Artificial: “Made by or resulting from art or artifice; contrived, compassed, or brought about by constructive skill, and not spontaneously; not natural. Artificial in result, as well as in process.” from the Oxford English Dictionary
Intelligence: “The faculty of understanding; intellect. Understanding as a quality of admitting of degree; superior understanding; quickness of mental apprehension, sagacity.” from the Oxford English Dictionary
Oxymoron? Yep, it’s one of those.
Seems like everybody’s going bonkers over chatbots, and not just the kiddies wanting to get one over on their teacher. CNN pontificators, internet influencers, and now even politicians are spouting wisdoms like they’re using GPT-4 for their media consultants. Given the current state of writing in the U.S., GPT-4 may be a godsend — overlooking the wooden Trojan horse in the room.
Here’s what I don’t understand: scraping a gazillion data points into a giant mosh pit, then pulling them out again, looking like a coherent thought — exactly how does that work, and what’s the point? As if we don’t have enough bad writing on the Internet? Research? Yeah, I’d love to find a decent research engine ever since Google went crazy selling ads, though I digress. Don’t be evil — though being insanely greedy is fine.
I would like to understand how chatbots work, but Reid Hoffman’s book didn’t get me there. I’m a slow reader — the denser the concept the slower I go — I was finished with Impromptu in an evening. There’s no there there to steal Gertrude Stein’s best line. Waste of a perfectly good tree to print it, and how many tons of West Virginia coal to generate the power?
The book was recommended by Rory Stewart, of The Rest Is Politics podcast fame among other achievements — such as resigning from Boris Johnson’s cabinet in protest against Johnson’s play on Brexit. I’d go hiking with Rory anytime, long as we don’t go to Afghanistan like he did. Stewart may hone his politics to a keen edge, but apparently doesn’t know diddly squat about AI. Nor do I, OK?
So I promptly emailed Bezos and ordered Impromtu, Amplifying Our Humanity Through AI authored by “Reid Hoffman with GPT-4.” Ah, so cute, listing his computer app as a co-author. I wonder if Rory was reading by candlelight in Jordon where he’s presently helping out and fell into a dark hole — just saying, Rory, just saying.
Because Hoffman must either a) be a shareholder in OpenAI, GPT-4’s parent company, b) just graduated from the Wharton School of Business where they teach management types like Donald Trump to do that time-honored practice of “Look over there! Look over there!” Or c) following in Sam Bankman-Fried’s footsteps. Hoffman has him some brass ones. But the best I can tell is that he knows no more than the rest of us about AI fundamentals, nor does he see anything possibly bad coming as a result.
It was said, the ugly secret about block chain technology underpinning Bitcoin et al. was the massive waste of energy sucked out of world to run the computers. There’s another massive energy suck going on related to AI, and if we don’t figure it out soon, we’ll be wishing we had.
“Look over there! Look over there!”
Impromptu is a long-ish magazine article that belongs in People magazine, some 240 pages of pablum, like another take on the famous Seinfeld show about nothing, only with no humor. Evidently, Reid Hoffman is known to the computer community, coming from LinkedIn fame.
In a recent Medium piece, John Gorman declared the ones who were busy last year shilling bitcoins are the ones hot to trot on how AI will change the world this year. I envy Gorman’s talent at yanking at a subject until you hear the throat rattles.
“Go through the ‘expert’s’ history — chances are if he was hella bullish on VR [virtual reality] in 2020, web3 [whatever happened to that?] in 2021, and the metaverse [Zuckerberg’s recent play] in 2022, the credibility just isn’t there. That’s just good-ol’ fashioned trend-hopping. Only certain types of people can afford to be spectacularly wrong that often for that long: deeply insecure men who accumulate their wealth by getting the rest of us to believe ‘this time it’s different’. [See: Cramer, Jim]” from All These Insufferable AI Bros Were CryptoBros Last Year, by John Gorman in Medium
I was giggling by the time I read the ‘see Cramer, Jim.’ It’s past time to nominate the Taxi script writers to the most quotable of all time.
Tom Friedman has made a name for himself keeping track of trends for the New York Times. He has an eye for speculating how the next new thing will affect civilization. Overlooking his cheerleading for the computer age — the COMPUTER AGE! — he does make interesting associations between new and old. Friedman’s gone quiet on the book publishing front of late, but he’s still on the Times payroll.
I respect Friedman’s eye enough to prod Google for his stories: We Are Opening the Lids on Two Giant Pandora’s Boxes is his NY Times opinion piece on the subject is worth reading — unlike Reid Hoffman’s. Friedman’s pandoras in question are AI and climate change, which are definitely not unrelated, given the joules AI consumes. I also noted the ever-grammatically correct headline plural ‘lids’ and ‘boxes.’
“Let’s face it, we did not understand how much social networks would be used to undermine the twin pillars of any free society — truth and trust. So if we approach generative A.I. just as heedlessly — if we again go along with Mark Zuckerberg’s reckless mantra at the dawn of social networks, “move fast and break things” — oh, baby, we are going to break things faster, harder and deeper than anyone can imagine.” from We Are Opening the Lids on Two Giant Pandora’s Boxes, by Thomas Friedman, NY Times
“Models like GPT-4 don’t do any one thing except predict the next word in a sequence.” from The Surprising Thing A.I. Engineers Will Tell You if You Let Them by Ezra Klein, NY Times
Later in the article, Klein delivers the best line on the subject: “it’s turtles all the way down.” Say amen.
So what are those chatbot predictions based on? A ginormous scrubbing of online sources, data mining of epic proportions via algorithms — plagiarism on steroids? Then grim little men in basement cages ‘train’ the app over and over again — and who trains the trainers? If you suspect facial recognition software misinterprets dark skin, who do you suppose didn’t train those apps?
Hoffman is too busy stroking his chatbot to explain — oh, was that HAL talking over the intercom? For those too young and needing a link, see 2001 Space Odessey.
I found one curious post on Medium: Book Review: Impromptu Amplifying Our Humanity Through AI by Reid Hoffman and GPT-4. The authors signed it “AI and Insights” — you can draw your own conclusions. Written by committee? Again spit balling, AI hasn’t gotten so far with writing catchy headlines. I get the impression GPT-4 did a jail break and signed onto Medium with its own membership. HAL, you calling?
Reid Hoffman has a ‘part’ on Education:
“Reid: In light of that, how would you recommend that American public education handle large language models such as you?
“GPT-4: I would recommend that American public education use tools such as me to augment and enhance the learning experience of students and teachers…” from Impromtu by Reid Hoffman.
Reid, can we talk? About that American public education: education can’t use jack, seeing as it isn’t sentient, not to mention oppossible thumbs. Neither you nor your chatbot knew that very basic language concept? Did your chatbot fail English 101?
Augment versus enhance. Class, please explain to Reid there’s no need to repeat the verb. Maybe some stoner trained his HAL to think two verbs doubled the emphasis. Smoke better weed, Reid.
I expect Reid was pure misery for his elementary school English teachers.
But if this kind of thing gets your blood flowing, then by all means by gosh and by golly, good gracious, yes, order a copy — order one for everyone in your family. Jeff is standing by at Amazon.
Search for AI on the Web and Python, the programming language, come back, along with C++ and Lisp from early programming days.
AutoCAD 2.53 was coded on Lisp in the 80s’ dark age , to give one example. I was very happy back then having software instead of drafting longhand. Ureeka! Even though it kept crashing the 286 IBM clone. Took decades and lots of blue screens of death to make AutoCAD work, and it stills sucks, but who’s complaining?
So what if the computers on the Apollo missions to the moon had a smaller processor and less RAM than an iPhone? The gods of chance were working. And John Glenn had real balls.
“Python is widely used for artificial intelligence, with packages for several applications including general AI, machine learning, natural language processing, and artificial neural networks. The application of AI to develop programs that do human-like jobs and portray human skills is machine learning. Both artificial intelligence and machine learning are closely connected and are being used widely today.” from Wikipedia page on Artificial Intelligence.
But an app to predict the next word, after scanning Facebook ditties, clever programming as that may seem, by definition (aka artifice) do not yield creative writing — more like computerized ventriloquism. Mathematicians consider a theorem that can be proved only 90% of the time fails on its face. But what’s the harm in a little wordplay? Ha, ha, good joke.
Reid: “So, GPT-4, please write an internet-length paragraph on how fantastically fabulous chatbots are.”
GPT-4 “Got you covered, Reid old buddy. Try this: Chatbots will save mankind. All you need is a honking big server farm and belief — oh, and buy my book! “Right, HAL?”
“HAL?”
Chatbots are the next new thing? Chatbots are to real intelligence as wind-up monkeys with cymbals are to Beyoncé, the American singer-songwriter, performing live. Machine learning? Machine parroting. And live parrots have mindfulness — overlooking the fact that only parrot-lovers want to kiss them beak to beak.
However, let’s take an example from my own profession: architecture. Suppose I need to program a public library, which can contain a wild permutation of functions, from reading spaces for parents to read to their toddlers to archival collections from the 18th century. I’ve designed libraries since the 80s — when people ask why, my response is because people still find them useful tools — and nice places to spend time.
Knowing what goes into a library, in order to design one, does not require knowing what the world’s libraries look like, though precedence is helpful. The great library in Alexandria, Egypt was said to contain the known world’s scrolls, though it left no records.
The gang that designed the new Alexandria Library spit-balled a design competition, a design that became a reality.
How’s that for a library form? I’ve only seen photos. My wife spent summers on a beach nearby, and still wants to return. Summer days on a beach were what her mother gifted her. The new library came years after.
Snøhetta, the Norwegian architects in question, also designed another cultural anachronism I’d love to visit — I’d even sit through Wagner’s Ring Cycle just to be there.
AI can’t duplicate even dullards like architects jotting notes listening to what local librarians and their patrons are seeking, and an understanding of the library board’s budget, the awareness of where this library might best fit in a downtown, and finding an appropriate form. Architects have two skills chatbots don’t: imaginations and skills at curation.
There was a professor at Yale, Allan Greenberg, who taught a course on classical architecture — a quiche course I avoided. It wasn’t so much I feared being proselytized to as I needed to spend my too-brief time in the Art & Architecture building on more meaningful subjects, such as dreaming of new buildings, learning Fortran IV and watching how my buddy Wong drew his beautiful ink-on-rice-paper axonometric drawings.
I have a hammer — you got a damn nail? Greenberg preached at his few disciples that classical architecture was the end-all, be-all style — to be clear, Western classical, not Moorish — and heavens not Edo-period Japanese. I will admit my profession has regularly indulged in hammering designs, but those have never been high points. Maybe a sly aside, wink wink say no more as the Monty Pythons would say.
In most cases, I can’t tell you where a building’s form springs from — the Muse is Steven Pressfield’s source — and I’ve waited many days at a drafting board for a Greek goddess to phone in. But whither it cometh, early in the process or very late in the day, the image is what is need. Without an overarching theme, valid in its own time and into the future, it’s a waste of bricks and mortar, metaphorically and physically.
Let’s suppose it’s 2050 and AI is in full throttle mode, replacing genius folks like stubborn architects. Only the rich will afford human architects because by then AI apps will be so much commodity. But the Republican Supreme Court AIs will be declaring the Founding Fathers — whoever they were — intended to write free enterprise into the Constitution, and the Progressive AIs will be complaining it’s pointed against the poor folk in Mississippi not being able to afford human architects, and the AFL-CIO will be striking in San Jose, by then a fully automated cyborg city. Do you suppose celebrity AIs will be yelling at each other, getting divorced, getting murdered?
Architects are generalists to our souls.
We need to juggle numbers of subjects just to wrestle a design down to its important parts. It’s a way of thinking, always needing to step back from the weeds to see the field — and waiting on the damn Muse. Lots of compromise, and little certainty. Though the day to day is spent accumulating the necessary data against which to make a judgement.
Therein lies where AI could contribute — not final synthesis so much as a resource for human judgement.
Will some inwardly focused Geni programmer dude wrap all this in a bottle? Very cool! But don’t sell off your bitcoins for it, ’cause it ain’t been invented.
“ ‘Automated systems should provide explanations that are technically valid, meaningful and useful to you and to any operators or others who need to understand the system, and calibrated to the level of risk based on the context,’ the blueprint says. White House’s Blueprint for an A.I. Bill of Rights
“Love it. But every expert I talk to says basically the same thing: We have made no progress on interpretability, and while there is certainly a chance we will, it is only a chance. For now, we have no idea what is happening inside these prediction systems. Force them to provide an explanation, and the one they give is itself a prediction of what we want to hear — it’s turtles all the way down.” from The Surprising Thing A.I. Engineers Will Tell You if You Let Them by Ezra Klein, NY Times
But how can you call these ‘prediction systems’ when all they can do is assemble other folks’ stuff from sources as profound as Twitter and Facebook? When was the last time you read a great quote on LinkedIn?
Being a sincere skeptic about collective human endeavors, I stand with Philip K Dick with his faith in the species’ talent for leaping like lemmings. So if Dick were still with us, his sci-fi plot might go like this: little Philip wakes up from a dream on his sixteenth birthday with total recall of all the words ever written in every language. He becomes the most famous genius ever, and the entire world’s population turns to him for leadership, only for their fearless leader to go insane for grasping the whole of our irreconcilable contradictions.
How, pray tell, does an earnest programmer (that’s right, clever but fallible humans with quirks like warts) train an algorithm absent the quirks? Then compound the problem with a gazillion iterations so even he can’t predict an outcome? God-like?
“But if A.I. gives us a way to cushion the worst effects of climate change — if A.I., in effect, gives us a do-over — we had better do it over right. That means with smart regulations to rapidly scale clean energy and with scaled sustainable values. Unless we spread an ethic of conservation — a reverence for wild nature and all that it provides us free, like clean air and clean water — we could end up in a world where people feel entitled to drive through the rainforest now that their Hummer is all-electric.” from We Are Opening the Lids on Two Giant Pandora’s Boxes, by Thomas Friedman, NY Times
I sure hope Reid Hoffman sold his stock in FXT before it folded. Though if he did, why is he writing drivel for new suckers?
Coda
While researching this topic, I came across research engines that may help writers struggling with getting their facts straight. In no particular order, with the bolded links back to Wikipedia:
