GPT-3 — let’s define it as the autocomplete tool by OpenAI trained on a large amount of uncategorized text from the internet — is quite impressive, comparable to what happened to AI image processing from 2012 onward. We can safely ignore the hype — it’s probably a dead end in terms of reaching artificial general intelligence (see its performance on the Turing Test). And I doubt it’s going to replace developers. But as an autocomplete, at guessing common sense or trivia questions, it’s a leap forward. Here I am asking Alexa for the third time to lower the volume, and this thing can almost handle a conversation.

Anyway because of how it works (you give it some text, a prompt, and it guesses what comes next) in recent weeks the internet got inundated with things they made GPT-3 do. There’s even a course in creative writing taught by GPT-3 (which is probably as valid as most creative writing courses).

Like all good AI GPT-3 never admits of not knowing an answer, it’d rather make up stuff, weird stuff sometimes, nonsense but nicely written nonsense. It might not make sense but at least it’s syntactically correct. It’s an idea machine, and a quite funny one. Here’s one of its replies, when prompted by Arram Sabeti to write an essay on human intelligence:

I propose that intelligence is the ability to do things humans do. [..] The brain is a very bad computer, consciousness is a very bad idea.

A talking mouth chanting algorithmically generated prayers. Given they’re nonsense to begin with, why not?

Toby Shorin:

In both the medieval and traditional forms of society, mankind was at the whim of God and nature. We could die in any number of ways. A locust wave (sent by God) ruined our crops, or you ate a poisonous berry, were bitten by a snake, or attacked by a bear, or some other cruel fate. To mitigate these eventualities the best we could do was pray to God, or for traditional humans, participate through ritual in the regeneration and renewal of the cosmos, an effort to help reinstantiate the natural order. So humans, weak and marginal, were just one figure in a cosmos of acting, agential beings and spirits.

Is this too not a form of full autonomy? Could we not say that human beings lived in a fully automated world then? One in which we were not at the center, but at the edges, just one part?

What if our desire for full autonomy is not a desire for “maximal mastery and total liberation, but the desire for limited agency? The desire to live once again in a naïve state of belief, one in which we are not paralyzed by optionality?

Arvind Narayanan:

I will focus the rest of my talk on this third category [predicting social outcomes], where there’s a lot of snake oil.

I already showed you tools that claim to predict job suitability. Similarly, bail decisions are being made based on an algorithmic prediction of recidivism. People are being turned away at the border based on an algorithm that analyzed their social media posts and predicted a terrorist risk. […]

Compared to manual scoring rules, the use of AI for prediction has many drawbacks. Perhaps the most significant is the lack of explainability. Instead of points on a driver’s license, imagine a system in which every time you get pulled over, the police officer enters your data into a computer. Most times you get to go free, but at some point the black box system tells you you’re no longer allowed to drive.

Il bot che genera farfalle

@mothgenerator è un bot di Twitter che genera sia le farfalle che i loro corrispettivi nomi in latino:

This bot tweets make-believe moths of all shapes, sizes, textures and iridescent colors. It’s programmed to generate variations in several anatomical structures of real moths, including antennas, wing shapes and wing markings.

Another program, which splices and recombines real Latin and English moth names, generates monikers for the moths. You can also reply to the account with name suggestions, and it will generate a corresponding moth.

Yuval Harari e l’uomo inutile

Ezra Klein di Vox ha intervistato Yuval Harari, autore di Sapiens e, più recentemente, di Homo Deus. Parlando di AI, Harari distingue fra intelligenza e coscienza — che vanno di pari passo negli umani, ma non per forza devono coesistere (o la seconda deve essere necessaria affinché la prima esista) in un’intelligenza artificiale:

Intelligence is not consciousness. Intelligence is the ability to solve problems. Consciousness is the ability to feel things. In humans and other animals, the two indeed go together. The way mammals solve problems is by feeling things. Our emotions and sensations are really an integral part of the way we solve problems in our lives. However, in the case of computers, we don’t see the two going together.

Over the past few decades, there has been immense development in computer intelligence and exactly zero development in computer consciousness. There is absolutely no reason to think that computers are anywhere near developing consciousness. They might be moving along a very different trajectory than mammalian evolution. In the case of mammals, evolution has driven mammals toward greater intelligence by way of consciousness, but in the case of computers, they might be progressing along a parallel and very different route to intelligence that just doesn’t involve consciousness at all.

Un passaggio importante del libro è la possibilità che, come conseguenza dell’automatizzazione, una larga fetta dell’umanità possa perdere la sua valenza economica (e di conseguenza politica) — ovvero diventare ‘inutile’ per lo stato e l’economia. Quando e se questo succederà, il sistema perderà anche l’incentivo di investire su questa classe di persone (la ragione per cui abbiamo università, assistenza sanitaria, etc. etc. è che queste cose ci rendono produttivi).

A questo punto che facciamo? Una possibilità spesso menzionata è che si finisca col nascondersi e col cercare di dare un significato alla propria esistenza tramite la realtà virtuale. Scenario triste, ma non nuovo, dice Harari: sono migliaia di anni che troviamo conforto, significato e modelliamo la nostra esistenza attorno a realtà virtuali che fino ad oggi abbiamo chiamato ‘religione’:

You can think about religion simply as a virtual reality game. You invent rules that don’t really exist, but you believe these rules, and for your entire life you try to follow the rules. If you’re Christian, then if you do this, you get points. If you sin, you lose points. If by the time you finish the game when you’re dead, you gained enough points, you get up to the next level. You go to heaven.

People have been playing this virtual reality game for thousands of years, and it made them relatively content and happy with their lives. In the 21st century, we’ll just have the technology to create far more persuasive virtual reality games than the ones we’ve been playing for the past thousands of years. We’ll have the technology to actually create heavens and hells, not in our minds but using bits and using direct brain-computer interfaces.

For voice to work really well you need a narrow and predictable domain. You need to know what the user might ask and the user needs to know what they can ask. This was the structural problem with Siri – no matter how well the voice recognition part worked, there were still only 20 things that you could ask, yet Apple managed to give people the impression that you could ask anything, so you were bound so ask something that wasn’t on the list and get a computerized shrug.

Conversely, Amazon’s Alexa seems to have done a much better job at communicating what you can and cannot ask. Other narrow domains (hotel rooms, music, maps) also seem to work well, again, because you know what you can ask. You have to pick a field where it doesn’t matter that you can’t scale.

Provate a ispezionare una delle immagini del vostro news feed del Facebook: molto probabilmente avrà un tag alt, automaticamente popolato con attributi che la descrivono – sole, montagna, natura, etc. a seconda di quale che sia il soggetto. Questi attributi non sono inseriti dagli utenti, ma automaticamente da Facebook, che analizza ciascuna immagine caricata e prova a comprenderne il contenuto.

Un’estensione per Chrome vi dà un’idea della quantità di informazioni che Facebook può ricavare con questa tecnica, permettendovi ti testarne le capacità su qualsiasi immagine.

Eliezer Yudkowsky risponde su LessWrong ad alcune domande riguardo all’automatizzazione e alla disoccupazione, due cose che (dice lui, riporto io, semplificando molto: leggetelo per intero) potrebbero essere correlate in un futuro (molto) distante – quando e se avremo un AI superintelligente –, ma che per il momento non sono causa ed effetto:


Many people would hire personal cooks or maids if we could afford them, which is the sort of new service that ought to come into existence if other jobs were eliminated – the reason maids became less common is that they were offered better jobs, not because demand for that form of human labor stopped existing. Or to be less extreme, there are lots of businesses who’d take nearly-free employees at various occupations, if those employees could be hired literally at minimum wage and legal liability wasn’t an issue. Right now we haven’t run out of want or use for human labor, so how could “The End of Demand” be producing unemployment right now? The fundamental fact that’s driven employment over the course of previous human history is that it is a very strange state of affairs for somebody sitting around doing nothing, to have nothing better to do. We do not literally have nothing better for unemployed workers to do. Our civilization is not that advanced.


Q. But AI will inevitably become a problem later?

A. Not necessarily.  We only get the Hansonian scenario if AI is broadly, steadily going past IQ 70, 80, 90, etc., making an increasingly large portion of the population fully obsolete in the sense that there is literally no job anywhere on Earth for them to do instead of nothing, because for every task they could do there is an AI algorithm or robot which does it more cheaply. That scenario isn’t the only possibility.

Harvard Business Review:

Technological revolutions tend to involve some important activity becoming cheap, like the cost of communication or finding information. Machine intelligence is, in its essence, a prediction technology, so the economic shift will center around a drop in the cost of prediction. The first effect of machine intelligence will be to lower the cost of goods and services that rely on prediction. This matters because prediction is an input to a host of activities including transportation, agriculture, healthcare, energy manufacturing, and retail.

When the cost of any input falls so precipitously, there are two other well-established economic implications. First, we will start using prediction to perform tasks where we previously didn’t. Second, the value of other things that complement prediction will rise.

Un esperimento di Google in grado di riconoscere quello che disegnate. Avete venti secondi di tempo per disegnare un concetto, e Google in quegli stessi venti secondi dovrebbe riuscire a identificare quel che state disegnando.

Ci ha preso quasi sempre con i miei decisamente orribili e confusi schizzi. Per intenderci: l’umano che siede di fronte a me ci ha preso meno di Google su quel che rappresentassero.

(Altri esperimenti inquietanti qui)

Un nuovo trailer di Lo and Behold, il documentario di Werner Herzog su internet e l’intelligenza artificiale. Ne attendo l’uscita trepidante.

S’intitolerà Lo And Behold: Reveries Of The Connected World; Verrà presentato al Sundance Festival, che inizia domani. Ci sarà anche Elon:

LO AND BEHOLD traces what Herzog describes as “one of the biggest revolutions we as humans are experiencing,” from its most elevating accomplishments to its darkest corners. Featuring original interviews with cyberspace pioneers and prophets such as Elon Musk, Bob Kahn, and world-famous hacker Kevin Mitnick, the film travels through a series of interconnected episodes that reveal the ways in which the online world has transformed how virtually everything in the real world works, from business to education, space travel to healthcare, and the very heart of how we conduct our personal relationships.


There are two main problems for any brain simulator. The first is that the human brain is extraordinarily complex, with around 100 billion neurons and 1,000 trillion synaptic interconnections. None of this is digital; it depends on electrochemical signaling with inter-related timing and analogue components, the sort of molecular and biological machinery that we are only just starting to understand.

Even much simpler brains remain mysterious. The landmark success to date for Blue Brain, reported this year, has been a small 30,000 neuron section of a rat brain that replicates signals seen in living rodents. 30,000 is just a tiny fraction of a complete mammalian brain, and as the number of neurons and interconnecting synapses increases, so the simulation becomes exponentially more complex—and exponentially beyond our current technological reach.

This yawning chasm of understanding leads to the second big problem: there is no accepted theory of mind that describes what “thought” actually is.

Un post da leggere per intero. Per il futuro prossimo, credo che possiamo stare tranquilli.