July 2023 - nexa - server-nexa.polito.it

“Emergence” isn’t an explanation, it’s a prayer. A critique of Emergentism in Artificial Intelligence

by Daniela Tafani

“Emergence” isn’t an explanation, it’s a prayer A critique of Emergentism in Artificial Intelligence <https://ykulbashian.medium.com/?source=post_page-----ef239d3687bf----------…> [cardboarddreams] Emergence is the notion that in a complex system, the interactions of the whole may exhibit properties that are not present in the individual parts. It is most often applied to examples in physics and nature, such as the collective behaviours of ant colonies, the self-organizing principles of social groups, or the macro properties of molecules. In the last few decades, it has given rise to emergentist perspectives of human cognition<https://onlinelibrary.wiley.com/doi/10.1111/j.1756-8765.2010.01116.x>, and even of consciousness<https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7597170/>. These are based on the recognition that the complexities and mysteries of the human mind, being a part of nature, may be characterized as emergent phenomena. This approach has an intuitive appeal. It is supported by the superficial facts: the brain — the source of intelligence and consciousness— is most certainly a complex of interconnected neurons. Emergentist interpretations of human behaviour may also boast some recent wins—the proliferation of LLMs (e.g. ChatGPT) may be seen as one such success. This has reignited the discussion of whether emergence is the best way to frame intelligence. You may have noticed that the last paragraph switched between two subtly different uses of “emergence”. The first use was to describe an observed emergent property; consciousness, we have seen it, likely emerges out of neuronal interactions. The second was to use it as a hypothesis to predict the outcome of a complex, unknown system, with the hope that a desired property will emerge; e.g. intelligence will arise from the interactions of artificial neurons at scale. The latter example is the driving motivation behind the multi-million dollar Human Brain Project<https://en.wikipedia.org/wiki/Human_Brain_Project>. The project justifies its cost by leaning on evidence from observed instances of emergence. Doing so paints emergence as a theory; that is, existing observations can be used to justify future predictions. But emergence is not a theory. Emergence can only be ascribed to a phenomenon in retrospect, once you already know what has “emerged”. The higher-level properties that emerge are qualitatively different from those at the lower-level — otherwise it wouldn’t be “emergence”. So by necessity they could not have been predicted from the lower-level ones. The properties of “intelligence” could not have been logically foreseen from the properties of neurons unless you had already observed that property emerge in a similar substrate. And even then it’s just a guess that is likely to be wrong given the complexity of the interactions involved; small differences can easily invalidate the hypothesis. In both cases emergence gives no new information: when explaining existing examples it gives you no new insights about the processes except that they happen; and when predicting unknown behaviours it gives very poor guarantees that anything you expect to happen will do so. Emergence is only really valid as a general metaphysical classification of certain phenomena. It’s a metaphysical category, like “cause”, “effect” or “change”. Using the word when explaining cognition is not wrong per se, it just has no real meaning or explanatory force. It’s like having a theory of “thing-happened-ness” — it’s correct, but void of content. Take, for example, the following quotes from a review article on emergence: This process gives rise to an emergent tendency to facilitate perception of items consistent with the patterns of English orthography, without explicitly representing this knowledge in a system of rules, as in other approaches. … However, such modes of thought themselves might be viewed as emergent consequences of a lifetime of thought-structuring practice supported by culture and education. Emergence in Cognitive Science, McClelland<https://onlinelibrary.wiley.com/doi/10.1111/j.1756-8765.2010.01116.x> If you removed the word “emergent” from the above two sentences, would anything important change? Indeed any sentence that includes “emergent” would give the same information if you removed it; “it gives rise to emergent properties” means the same as “it gives rise to properties”, or “there is an emergent tendency” is not substantially different from “there is a tendency”. Adding “emergent” to any sentence doesn’t increase its useful information content.¹ Emergence has no information that fundamentally differentiates it from a “miracle”. If I were to say that applying transformers<https://arxiv.org/abs/1706.03762> to Neural Networks creates intelligence through a miracle, I would be ridiculed. Were I to say that they create intelligence through emergent interactions, suddenly they gain an air of scientific credibility — but what have I added to the conversation with the use of that word? What quantifiable scientific facts are entailed in the term “emergent”? There are none. In cognitive science, emergence is regularly used to “explain” the connection between two phenomena, when it is otherwise complex and difficult to predict: e.g. how neuronal firing gives rise to consciousness, or transformers to the appearance of language comprehension. Where there may be a connection, but nothing more is known or can be proved, emergence is a placeholder that fills the gap. The word gives weight and gravitas to what is essentially a blank space. Despite emergence contributing nothing of substance to the discussion, as a concept it admittedly has a compelling intuitive appeal. There is a wonderful feeling about the notion of emergence. It does seem to be adding something valuable, as if you’ve discovered a magical ingredient by which you can explain mysterious phenomena. That’s the reason it continues to be popular, and gets inserted into scientific discussions. It convinces the listener that something has been explained with scientific rigour when all we’ve done is to say “it’s complicated”. Besides the good feeling, however, emergence is void of any explanatory power. And so it has no scientific value in a predictive capacity. You can’t use it to say anything about what an unknown system will do; only what you hope it will do. When applied to pie-in-the-sky AI futurism, emergence has become synonymous with “I’m sure the system will work itself out”. It indicates that the author has a feeling that a complex system will align at some point, but no clear sense of how, why, or when. Insofar as intelligence does manifest in a specific instance, “emergence” doesn’t tell us anything interesting about how it happened. And insofar as intelligence hasn’t yet manifested, emergence doesn’t tell us when it will or what direction to take to get there. In the field of AI development, emergence is invoked whenever someone encounters a phenomenon in the human mind and has no idea how to even start explaining it (e.g. art, socialization, empathy, transcendental aesthetics, DnD, etc). If said researcher already has a working theory of AI, this realization is disheartening. So they look deeper into the matter, find some point of overlap between the existing theory and the missing behaviour, and assume that with enough time and complexity the missing pieces will emerge. Emergence is attractive in such cases because it puts the author’s mind at ease, by making it seem like they have a viable mechanism that only needs more time to be vindicated. It placates their inner watchdog, the one that demands concrete, scientific explanations. Emergence, being related to complexity and superficially validated by experiments such as Conway’s Game of Life, is enough to lull that watchdog back to sleep. This justifies continuing to ignore any shortcomings in a theoretical model, and persisting on the current path. Like the proverbial man who searches for his lost keys under the lamplight, because that is where the light is, he hopes that with enough persistence his keys will “emerge”. The only other alternative is to admit failure, and to give up any hope of accomplishing what you want within this lifetime. Scientists, it seems, can have superstitions too. And emergence has a powerful narcotic effect: it feels so reasonable and credible on a gut level². There are many factors that prevent a given researcher from investigating emergence too deeply and realizing that it lacks any substance. First, there appears to be a lot of external evidence to back it up in the natural world. This, as was pointed out, equivocates between retrospective and prospective uses of the term, and so legitimate uses are being conscripted to justify the illegitimate ones. Secondly, the fact that emergence exclusively concerns itself with intractably complex systems means anything behind its curtain by definition can’t be studied. So it conveniently excludes itself from exactly that analysis which would reveal it to be hollow. In the end emergence isn’t an explanation; it’s an observation combined with a recognition of ignorance. Wherever emergence shows up there is an implicit acceptance that everyone involved is at a loss for how to approach the topic. It’s not that properties like intelligence won’t emerge from neural activity, it’s that emergence is a placeholder that justifies and promotes a lack of interest in exploring the details behind the connection. It discourages investigation. By invoking the term, we are merely thanking the nature gods for granting us this emergent property (aka property), and trying not to examine their gifts too profanely or with ingratitude. This impulse is understandable, since we don’t think we’ll discover an answer if we were to dig in. But we shouldn’t allow our insecurities to masquerade as science, or else they may become ingrained to the extent that they are difficult to uproot. A false answer stands in the way of a true one. ¹ This used to say ‘You can remove “emergent” from any sentence and it would mean the same thing’, but that has caused some confusion, so to clarify: the word “emergent” when used as an adjective doesn’t add new or useful information; you won’t know any more about the subject than you did before. ² A self-aware researcher should notice if they have a strong intuitive or emotional reason for holding on to the idea. If you ever feel that emergence is so self-evident that it can never be disproved, that should give you pause — perhaps you have strayed outside the bounds of scientific inquiry and into metaphysical expositions. Not that there’s anything wrong with the latter… https://ykulbashian.medium.com/emergence-isnt-an-explanation-it-s-a-prayer-…

July 19, 2023

LLM (di serie "A"): possibili in Italia/Europa? [Era: nexa Digest, Vol 171, Issue 53]

by Damiano Verzulli

Il 19/07/23 13:30, Giuseppe Attardi ha scritto: > [...] > Le risorse di calcolo per costruire GPT-3.5 sono stimate in 10^23 > FLOPS per un costo di centinaia di milioni di $ > Meta, per rilasciare i suoi modelli, ha costruito un Research > Supercluster con 10.000 GPU Nvidia, che secondo Yann LeCun è già in > overbooking. Leggo da una fonte terza (Wikipedia) che "Leonardo" [1] ormai ha quasi un anno, è costato 240M€ e di picco fa 250 petaFLOPS (aka: ~10^17), anche grazie ai suoi 13.824 GPU-core. Leggo da altra fonte terza (Top500 [2]) che attualmente (06/2023) risulta 4° al mondo, come potenza di calcolo. Leggo dal sito ufficiale [3] che: "Leonardo's main goals are [...] The computational power of Leonardo will boost scientific exellences and industrial strenght across Europe...." Non si parla di IA/ML, né si accenna agli LLM. Ma faccio comunque fatica ad immaginare che queste tonnellate di ferro *NON* possano essere utilizzate dalla comunita' della ricerca Italiana (...magari, in modo coordinato con gli altri paesi EU, dove "giocattoli" simili sono comunque presenti) a questo scopo. Ovviamente non mi aspetto che parcheggiati davanti al Tecnolopolo ci siano una fila di TAXI, pronti a scattare all'ordine di Cineca, per "prelevare" i ricercatori in giro per l'Italia al fine di portarli al Tecnopolo... per conoscere il giocattolo e iniziare ad usarlo. Certo: se il dottorando X, o l'assegnista Y (o anche il Ricercatore Z o il docente K) sentono il bisogno di avere del ferro sul pianerottolo di fianco al loro studio, in UNIV [come accade in UniPI, ad esempio, con i sistemi NVIDIA qui discussi, qualche giorno fa]... allora il discorso cambia... Un'ultima nota a chiusura: sono cosciente che fra 10^17 e 10^23 c'e' *MOLTA* differenza (a proposito: qual'e' la fonte di 10^23?). Prima di preoccuparmi di questo, pero', attenderei di vedere che quei 10^17 stiano lavorando almeno come 10^16 per un buon periodo di ore/mese. Dopodiché sarei pronto ad alzare la mano e chiedere qualcosa... di piu' performante. Un saluto, DV [1] https://en.wikipedia.org/wiki/Leonardo_(supercomputer) [2] https://www.top500.org/lists/top500/2023/06/ [3] https://leonardo-supercomputer.cineca.eu/ -- Damiano Verzulli e-mail: damiano(a)verzulli.it --- possible?ok:while(!possible){open_mindedness++} --- "...I realized that free software would not generate the kind of income that was needed. Maybe in USA or Europe, you may be able to get a well paying job as a free software developer, but not here [in Africa]..." -- Guido Sohne - 1973-2008 http://ole.kenic.or.ke/pipermail/skunkworks/2008-April/005989.html

July 19, 2023

Re: [nexa] AI and antitrust in 10 minutes

by Giuseppe Attardi

> On 19 Jul 2023, at 07:21, nexa-request(a)server-nexa.polito.it wrote: > > Date: Wed, 19 Jul 2023 13:17:32 +0200 > From: Fabio Alemagna <falemagn(a)gmail.com> > To: Guido Vetere <vetere.guido(a)gmail.com> > Cc: Daniela Tafani <daniela.tafani(a)unipi.it>, > "nexa(a)server-nexa.polito.it" <nexa(a)server-nexa.polito.it> > Subject: Re: [nexa] AI and antitrust in 10 minutes > Message-ID: > <CACGmXuO94NnbKmRO3zqqazLNWOcOFry3=p-MFocTjodu1PmcLw(a)mail.gmail.com> > Content-Type: text/plain; charset="utf-8" > >> Il giorno mer 19 lug 2023 alle ore 12:51 Guido Vetere >> <vetere.guido(a)gmail.com> ha scritto: >> >> per quel poco che ho studiato la questione del cross-lingual transfer learning, la cosa riguarda il trasferimento da lingue con molte risorse (leggi: inglese) a lingue con scarse risorse >> c'è tutta una questione di relativismo culturale nella quale ora non è il caso di addentrarsi, ma in tutti i casi vale il discorso del downsizing: è difficile - per dire - che sia necessario trasferire qualcosa di urdu in italiano, quindi basterebbero coppie inglese-x > > Se la produzione letteraria in un dato linguaggio è scarsa, viene da > sé che il contributo ai parametri del modello sarà anch'esso "scarso". > Non vedo dunque il problema. Non è detto, ci sono tecniche di ribilanciamento. — Beppe

July 19, 2023

Re: [nexa] nexa Digest, Vol 171, Issue 53

by Giuseppe Attardi

> On 19 Jul 2023, at 06:51, nexa-request(a)server-nexa.polito.it wrote: > > Date: Wed, 19 Jul 2023 12:05:12 +0200 > From: Fabio Alemagna <falemagn(a)gmail.com> > To: Guido Vetere <vetere.guido(a)gmail.com> > Cc: Daniela Tafani <daniela.tafani(a)unipi.it>, > "nexa(a)server-nexa.polito.it" <nexa(a)server-nexa.polito.it> > Subject: Re: [nexa] AI and antitrust in 10 minutes > Message-ID: > <CACGmXuPNy4Y9uFBdL7O=QsCYtNYE3hP5gLbg-QzUZW=fPr8qKA(a)mail.gmail.com> > Content-Type: text/plain; charset="UTF-8" > >> Il giorno mer 19 lug 2023 alle ore 10:44 Guido Vetere >> <vetere.guido(a)gmail.com> ha scritto: >> >> un piccolo commento a caldo dopo aver dato una scorsa a questo illuminante intervento >> noi diamo per scontato che i LLM non possano che essere ciò che oggi ci viene proposto dal dupolio Microsoft \ Google > > Non mi pare esista al momento un duopolio riguardo gli LLM: ne > esistono decine di completamente open source, prodotti un po' in tutto > il mondo. Dipende da cosa consideri Large. I veri LLM, quelli la cui dimensione consente l’apparire di emergent abilities, solo pochi si possono permettersi di costruirli. E le dimensiini dei LLM sono finora cresciute esponenzialmente. D’altra parte, non avrebbe senso che migliaia di ricercatori chiedessero di fermare lo sviluppo di LLM più potenti di GPT-4, se questa non fosse il percorso di sviluppo più promettente. I LM cosiddetti Open Source (ma non è di source che si parla, ma dei parametri del modello), sono circa un ordine di grandezza più piccoli di quelli più grandi. Questo si ripercuote sulle loro capacità. Non bisogna farsi illudere dalle dichiarazioni degli sviluppatori nel confronto con altri LLM. I confronti vengono fatti su task specifici, su cui quei modelli sono ottimizzati. Ma i LLM contengono una mole superiore di conoscenze, tali che possono essere utilizzati per altri task, solo col prompting, senza fare fine-tuning. E il fine-tuning di un modello da 60-80 miliardi di parametri richiede comunque un server con almeno 4 GPU (altrimenti non sta in memoria) e diversi giorni di calcolo. Il risultato è spesso inferiore a quello di un LLM. Lo so per esperienza diretta personale e di altri. Infine, non vorrei lasciare a quei pochi che se possono permettere, le scelte su come fare un LLM e dovermi limitare a quello che loro graziosamente, o pelosamente per conquistare quote di mercato, mettono a disposizione. Vorrei poter avere la libertà di esplorare anche nuove strade. Anche solo per fare, come dice Vetere, modelli per la mia lingua, o per un settore specifico (salute, energia), o per determinati punti di vista (politici, economici, sociali, personali). > Anche il Technology Innovation Institute dell'Arabia Saudita > ha rilasciato un LLM come Open Source: https://falconllm.tii.ae/ The model uses only 75 percent of GPT-3’s training compute, 40 percent of Chinchilla’s, and 80 percent of PaLM-62B’s Ossia, forse gli arabi hanno i soldi per pagarsi le risorse computazionali per costruirsi un loro LLM, ma difficile che ce li abbiano i ricercatori europei, quando i progetti europei su AI dispongono di un centinaio di milioni in tutto per dozzine di progetti triennali con dozzine di partner. Le risorse di calcolo per costruire GPT-3.5 sono stimate in 10^23 FLOPS per un costo di centinaia di milioni di $. Meta, per rilasciare i suoi modelli, ha costruito un Research Supercluster con 10.000 GPU Nvidia, che secondo Yann LeCun è già in overbooking. Musk, mentre chiede di fermare lo sviluppo di LLM, ha ordinato anche lui 10.000 GPU per X.AI. Le startup come Converse.AI e Anthropic AI, hanno raccolto finanziamenti da 1-3 miliardi$, principalmente per comprarsi le risorse di calcolo. Il massimo che abbiamo in Europa è Mistral, con 100 milioni di VC. — Beppe

July 19, 2023

Re: [nexa] AI and antitrust in 10 minutes

by Fabio Alemagna

Il giorno mer 19 lug 2023 alle ore 12:51 Guido Vetere <vetere.guido(a)gmail.com> ha scritto: > > per quel poco che ho studiato la questione del cross-lingual transfer learning, la cosa riguarda il trasferimento da lingue con molte risorse (leggi: inglese) a lingue con scarse risorse > c'è tutta una questione di relativismo culturale nella quale ora non è il caso di addentrarsi, ma in tutti i casi vale il discorso del downsizing: è difficile - per dire - che sia necessario trasferire qualcosa di urdu in italiano, quindi basterebbero coppie inglese-x Se la produzione letteraria in un dato linguaggio è scarsa, viene da sé che il contributo ai parametri del modello sarà anch'esso "scarso". Non vedo dunque il problema. > per il resto: certo che tutti hanno il diritto di farsi un language model, il tema è quello del sostegno a questo tipo di sviluppi in ottica 'sociale' Non parlavo del diritto a farsi il proprio LLM, ma della possibilità di utiilizzare un LLM nella propria lingua. In altre parole, ChatGPT è accessibile a una vastissima platea mondiale proprio in funzione del numero della grossa quantità di lingue che parla. E comunque, non sottovaluterei la rilevanza dell'Urdu.

July 19, 2023

Re: [nexa] AI and antitrust in 10 minutes

by Guido Vetere

per quel poco che ho studiato la questione del cross-lingual transfer learning, la cosa riguarda il trasferimento da lingue con molte risorse (leggi: inglese) a lingue con scarse risorse c'è tutta una questione di relativismo culturale nella quale ora non è il caso di addentrarsi, ma in tutti i casi vale il discorso del downsizing: è difficile - per dire - che sia necessario trasferire qualcosa di urdu in italiano, quindi basterebbero coppie inglese-x per il resto: certo che tutti hanno il diritto di farsi un language model, il tema è quello del sostegno a questo tipo di sviluppi in ottica 'sociale' G. On Wed, 19 Jul 2023 at 12:10, Fabio Alemagna <falemagn(a)gmail.com> wrote: > Il giorno mer 19 lug 2023 alle ore 12:05 Fabio Alemagna > <falemagn(a)gmail.com> ha scritto: > > Anche il Technology Innovation Institute dell'Arabia Saudita > > ha rilasciato un LLM come Open Source: https://falconllm.tii.ae/ > > Pardon, Emirati Arabi. >

July 19, 2023

Re: [nexa] AI and antitrust in 10 minutes

by Giuseppe Attardi

Magistrale. In particolare sono convinto della necessità di scoraggiare il business model dei servizi gratuiti finanziati con la pubblicità, introducendo una tassazione altamente progressiva dei ricavi pubblicitari, come propone il nobel Paul Romer. So I am actually quite favourable, […] to a digital ad tax that creates more openness for alternative business models based on things like Wikipedia or subscription models in the online space. — Beppe > On 19 Jul 2023, at 04:44, nexa-request(a)server-nexa.polito.it wrote: > > From: Daniela Tafani <daniela.tafani(a)unipi.it> > To: "nexa(a)server-nexa.polito.it" <nexa(a)server-nexa.polito.it> > Subject: [nexa] AI and antitrust in 10 minutes > Message-ID: <08789adc08254f80af9be9f171ad7521(a)unipi.it> > > Daron Acemoglu > Thank you, Cristina, for that wonderful introduction and to you and Tommaso for inviting me. […] > As Cristina said, I'm going to talk about something that's partly inspired by my book. AI and > antitrust in 10 minutes. So that's a tall order especially if I try to blend in ideas from the book, so > let me jump into it. I'm going to do 10 question and answers in 10 minutes, but since that's a very > short time I'll just give you the answers. I'll let your imagination do the job of the what the questions > might have been to which. These are the answers. > …

July 19, 2023

Re: [nexa] AI and antitrust in 10 minutes

by Fabio Alemagna

Il giorno mer 19 lug 2023 alle ore 12:05 Fabio Alemagna <falemagn(a)gmail.com> ha scritto: > Anche il Technology Innovation Institute dell'Arabia Saudita > ha rilasciato un LLM come Open Source: https://falconllm.tii.ae/ Pardon, Emirati Arabi.

July 19, 2023

Re: [nexa] AI and antitrust in 10 minutes

by Fabio Alemagna

Il giorno mer 19 lug 2023 alle ore 10:44 Guido Vetere <vetere.guido(a)gmail.com> ha scritto: > > un piccolo commento a caldo dopo aver dato una scorsa a questo illuminante intervento > noi diamo per scontato che i LLM non possano che essere ciò che oggi ci viene proposto dal dupolio Microsoft \ Google Non mi pare esista al momento un duopolio riguardo gli LLM: ne esistono decine di completamente open source, prodotti un po' in tutto il mondo. Anche il Technology Innovation Institute dell'Arabia Saudita ha rilasciato un LLM come Open Source: https://falconllm.tii.ae/ > ma se ci pensiamo un attimo, questa necessità non esiste: forse si tratta solo di un'illusione propagandistica > perché dobbiamo dare per scontato che chiunque sulla faccia della terra abbia bisogno di generare testo in qualsiasi lingua? Esistono centinaia di lingue sulla faccia della terra, mi pare sia abbastanza scontato che chiunque possa avere il diritto di generare testo nella propria lingua. > a me ad esempio un LLM in italiano e inglese andrebbe più che bene, dunque sono sicuro che la maggior parte dei millemila miliardi di parametri di GPT4 non li userò mai (pur pagandoli) La capacità generativa di un LLM in una specifica lingua non dipende solo dal training effettuato su quella specifica lingua, ma anche da tutte le altre lingue. I concetti appresi in ognuna delle lingue su cui il LMM è stato allenato entrano a far parte della stessa rete neurale, e di conseguenza contribuiscono alla generazione. Non è molto diverso da quello che accade quando io leggo qualcosa in inglese e poi uso ciò che ho appreso da quella lettura per scrivere un saggio in italiano. > il fatto che lo stesso LLM debba servirmi per la generazione di testo e di software è anche abbastanza strano: sarei ben disposto a ricorrere a piattaforme diverse > insomma io vedo grandi possibilità di downsizing "by task" Un training "generalista" consente al LLM di poter fare collegamenti interdisciplinari. È già previsto che, successivamente al training generalista, si possa poi proseguire con un training "verticale" nel settore preferito, eventualmente "alleggerendo" il modello di parti non rilevanti in modo da aumentarne la velocità durante le inferenze e diminuirne i consumi. > e qui viene la pregnanza di ciò che Acemoglu dice: dovremmo, con le politiche pubbliche, incentivare uno sviluppo che vada in una diversa direzione, non solo cercare di vincolare la direzione monopolistica attuale > di fatto, nella ricerca di nuovi modelli, già si muovono diverse realtà pubbliche e private > G. > > > > > On Wed, 19 Jul 2023 at 10:12, Daniela Tafani <daniela.tafani(a)unipi.it> wrote: >> >> Daron Acemoglu >> Thank you, Cristina, for that wonderful introduction and to you and Tommaso for inviting me. […] >> As Cristina said, I'm going to talk about something that's partly inspired by my book. AI and >> antitrust in 10 minutes. So that's a tall order especially if I try to blend in ideas from the book, so >> let me jump into it. I'm going to do 10 question and answers in 10 minutes, but since that's a very >> short time I'll just give you the answers. I'll let your imagination do the job of the what the questions >> might have been to which. These are the answers. >> - Yes, generative AI has great potential, so I am completely convinced that this is a very >> interesting technology that can bring lots of goods and has capabilities so we can build on >> that. But I think let's move forward. >> - And yes, I believe that monopoly is everywhere in the tech sector. So here, perhaps I differ >> from many IO economists, and I subscribe to the duck test. If something looks like a duck, >> walks like a duck, and quacks like a duck, it is a duck. So if you have companies that have >> reached sizes that have never been in human history, and that dominate a particular line >> of business, they are monopolies. So we have to grapple with that. And that means all sorts >> of regulatory tools have to be considered, including antitrust. So this is absolutely on >> target. >> - And yes, in my view this is getting worse with foundation models. Because there is a >> likelihood that we may go towards a duopoly. With Microsoft Open AI and Google as the >> two key players, even though open source and many other competitors are going to try to >> get into foundation models, but the current business model of foundation models is very >> resource intensive. So that raises the possibility, does not in any way creates a certainty, >> but it raises the possibility. These two companies and their models are going to be the >> dominant ones on which many others will have to build, raising all of the issues of vertical >> product creation and all sorts of other questions that are going to be central for >> policymakers and economists to grapple with. >> - But no, I actually don't think monopoly power leading to high prices is the main problem >> that we're dealing with. You know, of course, that is a problem. But if the only issue was >> that because the foundation models are controlled by, you know, Google and Microsoft, >> they’re going to charge higher prices, and as a result, the apps that are developed on them >> are going to be more expensive, that would be, of course a pity and it's something we can >> do something about, but it wouldn't be the end of the world. So we get many new apps. >> They cost a little bit more. We don't get quite the consumer surplus. Woe is us, but not the >> end of the world. The problem is the direction of innovation. The problem is that the current >> market structure is selecting a particular direction of innovation. And that has much more >> sweeping consequences. Taking the set of products and technologies as given and pricing >> them above marginal cost and thus losing some of the welfare triangle is not the main issue. >> There’s the potential for doing much greater damage. No, this is not because of existential >> risk. In fact, like Cristina was implying, when all of these tech leaders are talking about >> existential risk I see it as either a blind spot or a ploy for making us not worry about the >> bigger risks. The bigger risks in my mind are in the labour market. Most of us earn our living >> in the labour market, so what happens to jobs is the most important issue. And the current >> direction of AI looks like it is going to follow some of the trends we have seen with digital >> technologies before. Failing to create the complementarities which human workers and >> skills, and instead going much toward much more towards automation, hence generating >> inequality, potential job losses, especially for workers without very specialised skills such >> as those with postgraduate degree. >> - And no, it's not just economics. There is a real danger here that the current direction of >> generative AI could again continue existing trends that we saw in social media degrade >> political conversations. Increase the amount of misinformation and disinformation, with a >> much more powerful tool. Create a particular type of ecosystem in online forums where >> people are drawn on the basis of emotion rather than engagement. And hence generally >> act towards the exploitation of people in their capacities and duties as democratic citizen. >> It is this twin: Inequality and Elimination of good jobs in the labour market, as well as >> erosion of democratic capacity that I think are most problematic. >> - No, I am not a Luddite. So I am not saying that this is in the nature of technology, nor that >> we should oppose technological change. The issue here is that we are not along the right >> path. What's great about technology in general and generative AI in particular, that it's a >> very highly malleable type of technological platform or what some economic historians >> used to call general purpose technology, meaning that you can use it for creating many >> apps, many different types of sub technologies and many different directions are possible. >> It is not complete idle talk. When people used to talk about social media and other online >> tools creating new democratic spaces today, it looks like very naive. When people in the >> 2000s said oh, online communication and social media are going to democratise >> communication. But that potential was there, and that potential is much greater with AI. >> When some people in the tech industry talk about generative AI being useful to humans in >> terms of getting better information, performing better tasks so generative AI… actually, I >> think the great potential that I mentioned at the beginning is precisely in being a human >> complementary technology. The tragedy of our current age is that we have almost all >> information that is at least codified available in some form. But we do not have the >> processing power to decide which one we should retrieve, how we should interpret how we >> should process, and which types of information we should engage with in different forms. >> Generative AI has the capability to improve human interaction with information and hence >> generate a lot of tasks, not just for knowledge workers, but for electricians, for carpenters, >> for educators, for healthcare workers. So that possibility is there. >> - But no, we are not going in the right direction. So we do need a redirection of technological >> change. >> - And no. I don't think it is naive or unrealistic to think about the redirection of technological >> change. One view which is common among some economists and some tech leaders goes >> back to either to the view that technology somehow has a preordained path and we just >> have to follow it. No, I am denying that and I think history is quite a good guide on showing >> how malleable technology is. Goes back to a saying by Ferdinand de Lesseps of fame, from >> the Suez and the Panama Canal, which we discussed in this book where he said, don't worry, >> men of genius would arise and solve all problems. Who are today's men of genius. Maybe >> some Altman or Elon Musk. But no, I don't think we should trust them. So I think the >> direction of technology is malleable, but it's also a societal choice. >> - And yes, as Cristina was hinting, antitrust has a very important role in this. For two reasons. >> One, because if we want alternatives. They are not very likely to come from a duopolistic >> or highly oligopolistic structure, especially one further empowered by killer acquisitions >> and these companies being a block other types of technologies that do not fit well with >> their business model. So if we want more alternatives that go more in a human >> complementary direction or more pro democratic direction, or create a more open >> competitive environment, I think we have to use antitrust tools including potential breakup >> of the largest companies which are too big and one other reason is because this type of >> power comes with enormous social power. And by that social power, I mean economic >> power, and also general social power. Tech companies have an enormous sway on public >> opinion, which I think is associated with their mega profits. And again, I don't think that >> creates a healthy environment. >> - And no, finally, I don't think antitrust is the main tool as Cristina was also hinting in her >> introductory comment. I think antitrust is a very blunt tool and I think for the redirection >> of technology we need a suite of tools which should include exactly how data is used and >> accessed. We need a new interoperability type of approach as well as how do we >> compensate and how do we actually encourage more creation of creative data. We also >> need to provide explicit incentives, such as, for example, those that have been successful >> in the field of renewable technology, where we encourage more of the socially valuable >> types of technologies. And I think there are a number of tools for that and we may also >> need more tax policies to discourage the worst types of business models and create room >> for alternative business models. So I am actually quite favourable, although I think much >> more study is needed, to a digital ad tax that creates more openness for alternative >> business models based on things like Wikipedia or subscription models in the online space. >> - Finally, I think we also need to rethink other tools that we have, like fiscal tools that >> currently create a very asymmetric playing field between capital and labour and going back >> to the job market and labour market inequality issues. I think equating marginal tax rate >> between capital and labour are things that we should definitely revisit. Thank you. >> >> https://mailchi.mp/cepr/central-bank-communication-rpn-seminar-series-516707 >> >> _______________________________________________ >> nexa mailing list >> nexa(a)server-nexa.polito.it >> https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa > > _______________________________________________ > nexa mailing list > nexa(a)server-nexa.polito.it > https://server-nexa.polito.it/cgi-bin/mailman/listinfo/nexa

July 19, 2023

DPA norvegese blocca trattamento di Meta

by Stefano Quintarelli

The Norwegian Data Protection Authority imposes a ban on Meta carrying out behavioural advertising based on the surveillance and profiling of users in Norway. The ban will initially apply until October … In December last year, the Irish Data Protection Commission issued a decision on behalf of all data protection authorities across the EEA which established that Meta has conducted illegal behavioural advertising. Since then, Meta has made certain changes, but a fresh decision from the Court of Justice of the European Union (curia.europa.eu) has stated that Meta’s behavioural advertising still does not comply with the law. Therefore, the Norwegian Data Protection Authority is now taking action by imposing a temporary ban. https://www.datatilsynet.no/en/news/aktuelle-nyheter-2023/temporary-ban-of-…

July 19, 2023