Going Infinite – Michael Lewis – 254 pages
Michael Lewis has become a literary superstar specializing in a unique genre of reportage -- the psychological / financial / cultural post mortem. He published his first book Liar’s Poker in 1989 and it provided insights into the selective recruiting process adopted by Wall Street that created the original incarnation of what we would now term the “bro culture” of Wall Street. That culture, a combination of mathematical expertise, extreme risk-affinity and aggression, was instrumental in fueling widespread abuses in the junk bond market. The book described Lewis' experiences that started in 1984 and those patterns were already linked to infamous cases like those of Ivan Boesky and Michael Milkin before he even published it in 1989. The book didn’t explicitly forecast the round of “innovations” in financial engineering that resulted in the 2008 meltdown but reading Liar’s Poker NOW makes it clear how the testosterone-fueled, high-risk “bro culture” hired in the mid-1980s rose up to senior management and enabled exponentially more dangerous tactics that made the entire mortgage-backed security disaster not only likely but inevitable.
Since that first book thirty four years ago, Lewis has developed a seemingly unerring sense of where to dive next to find the next interesting story on how quirky personalities, unique insights, emerging market trends and good old fashioned greed and corruption manage to combine to produce the next management or economic meltdown. Going Infinite, just published on October 3, 2023, fits perfectly in the larger Lewis canon in terms of the subject matter and Lewis’ unique approach to gaining insight and telling the story. The book is highly recommended for its ability to convey the day to day vibe of working with the people at FTX and Alameda Research and how that vibe led to its collapse. Unfortunately, the book also fails at times to appropriately tie together scattered facts in the book into useful insights or states conclusions which partially conflict with his own exposition. Perhaps most curiously, the book misses a chance to close the circle from Liar's Poker to the present by identifying a common theme – attempts to leverage (pronounced "exploit") people with extreme but narrow talents within systems designed to magnify miniscule discrepancies and convert them into huge profits while ignoring crucial shortcomings in those same resources that create the potential for great harm, both to those individuals and society.
Michael Lewis as Forrest Gump
As previously mentioned, Michael Lewis has an uncanny knack for being near interesting stories and extracting riveting tales from them just as they come into focus for the entire world. How did he do it this time? A friend of his was considering a substantial private equity investment in FTX in late 2021. The friend had attempted to gain a sense of what Sam Bankman-Fried (SBF), the CEO of FTX (and essentially its sister firm Alameda Research) was like but could not find anyone that could convey an accurate sense of SBF as an executive or an individual. The friend asked Michael Lewis if he could arrange a meeting with SBF to get a better handle on his demeanor, his mien. Lewis agreed, contacted SBF (somehow) and arranged an initial meet and greet. From that initial conversation, Lewis gave his own friend an enthusiastic thumbs up but also established an ongoing relationship with SBF that led to ongoing in-person conversations from that point through the day of SBF’s ultimate arrest in his condo in the Bahamas in November of 2022.
It's worth summarizing this information... Michael Lewis met Sam Bankman-Fried at the request of a friend to conduct (essentially) due diligence on Sam Bankman-Fried as a person and leader of a business venture. After a single meeting and a walk at a local park, Lewis gave his friend an enthusiastic thumbs up to do business with SBF (though Lewis does not state if the friend ultimately invested or stayed out). Lewis' total exposure to SBF was from late 2021 to SBF's arrest on November 14, 2022, essentially the last year of the operation of FTX and Alameda Research.
The Bob Problem and the Neurodiverse
The book devotes an entire chapter entitled How to Think About Bob to providing background on a crucial aspect of the core thinking process of SBF in particular but of many “neurodiverse” people more generally. In explaining his general thinking pattern to the author, SBF described a scenario in which your best friend Bob attends a dinner party with 99 other people and one of those other guests winds up dead. Police have no DNA evidence, the home where the dinner party took place was sealed so the only thing that is known is that one of the remaining 99 killed #100. That means your best friend Bob has SOME chance of being a murderer. So how do you change your behavior regarding your best friend?
As the author recounts it, SBF’s approach is to avoid insisting on absolutes (“my friend COULD be a killer so begin avoiding him entirely” versus “I trust my friend, he couldn’t POSSIBLY be the murderer, make no changes”). Instead, model the problem, calculate the probabilities of the outcomes, identify the value of the consequences, multiply probability x value for each possibility, then use the immediate scenario at hand to pick a path and move on. If Bob wants you to join a conference call to explain something to a client, sure, no problem. If Bob wants to bring the wife over for a BBQ on Saturday with the kids, maybe not. Lewis seems to recognize there are limitations to when this approach can be effectively used but doesn't elaborate on what those might be or how SBF and his employees made such distinctions or ever tried making such distinctions.
In high stakes business situations, especially those with a fiduciary responsibility regarding Other People’s Money or Other People's Lives, this approach is highly inappropriate. The book includes several incidents where SBF or coworkers identified GLARING shortfalls in accounts and security flaws in their systems involving hundreds of millions or even BILLIONS of dollars that should have triggered “all hands on deck” investigations to identify, resolve and correct. Instead, this tendency to “quantify” and compartmentalize risk consistently led the entire firm to downplay and dismiss issues affecting their clients – issues whose magnitudes and the lack of diligence to correct reflect criminal incompetence.
The larger problem for the author is that by failing to address the consequences of this mindset in the case of FTX, Lewis also failed to truly define its origins, its mechanics and its implications in a larger society. Missing that connection seems odd for an author who caught on to an earlier generational version of the same problem in his first book. With the understanding and vocabulary of the 1980s, Liar’s Poker explicitly identified a process by which Wall Street firms were purposely recruiting a specific combination of mental ability and personality they could further train to essentially act as pit bulls on trading floors and (later) computerized trading exchanges. The goal at that time was to quantify previously opaque risks then make outsized bets to extract outsized profits from short term distortions in securities markets.
In the vocabulary of the 1980s, Lewis characterized these candidates as mathematically gifted but reflecting a very aggressive, arrogant, sexist frat boy attitude – a stereotype that today pervades both the financial industry and software industry as “bro culture.” Leadership at Wall Street firms in the 1980s already had the frat boy underpinning from the 1970s generation of new hires but recognized the value of leveraging “quants” so they began recruiting “quants” who were otherwise just like them – frat boys. 1980s Wall Street wasn’t worried about sexual harassment and #metoo blowback, they just wanted these recruits to constantly scan the markets for opportunities to arbitrage and exploit to create profits from nothing.
After nearly forty years, the “class of 1984” now runs Wall Street. Over that same period, concepts about learning and education have provided a new vocabulary to describe characteristics and tendencies of people who might have been seen as dysfunctional in the past. Terms like neurodiverse, ADHD, Asperger’s Syndrome, etc. have grown commonplace and companies looking for an edge in highly technical, highly computerized sectors have discovered value in people further away from "normal" on the neurotypical / neurodiverse gradient. SBF himself was recruited for an internship and selected – not because he arrived with a polished resume, a clean suit, an airtight "elevator speech" and an understanding of stock trading, but because he scored highly in special games devised to identify "game playing talent."
People with mathematical or engineering backgrounds have obviously spent more of their formal education completing coursework that emphasizes physical phenomena or business processes into building blocks interconnected by inputs and outputs. Those blocks and connections can directly map to mathematical models of variables, coeeficients, etc. Present such people nearly any system and some portion of their brain will immediately launch a thread that starts decomposing the system into components, connections between those components representing inputs and outputs and factors indicating how much “signal” gets from one box to another. It’s what they do… For most neurotypicals with a bent towards math or engineering, this “background thread” might only consume two percent of their “CPU” and they can quickly set aside even that distraction, engage in the “present” and, you know, interact like, well… a neurotypical person.
For those further on the scale away from neurotypical, the distraction factor of this background thread might be five or ten percent. And frankly, they might be better at that instant modeling than a neurotypical and they might ENJOY spending time just contemplating that system they just devised in their head. For those even further away from neurotypical, this “background processing” becomes another input into what can become ADHD and – again, because they are actually GOOD at it – it becomes more enjoyable than other processing tasks they may not be so good at or enjoy. At that point, this type of constant “system analysis” and optimization becomes a coping mechanism and can morph into debilitating distraction.
Imagine someone at this point on the neuro spectrum needing to drive from their house to a business located to their west which can be reached by going counterclockwise from Street A to Street B then Street C or they can drive clockwise from Street A to Street D to Street C. The CCW route has a slower speed limit but fewer lights but has a hill enroute. The CW route has more lights, but faster speed limits and no hills. Which way should they go? An “average” person will flip a mental coin, pick a route and think about other things. An “average” engineer might contemplate these variables, think through if they want to optimize for time, distance or miles per gallon while looking for their car keys, then still flip a mental coin and drive. Someone further on the spectrum will think about their model and how to optimize the route the entire time they are driving to the business and back. Someone WAAAAY on the spectrum might become nearly paralyzed and avoid making an actual decision entirely.
In software engineering, there’s a term for this… Premature optimization. Developers who possess an understanding of a wide range of components in the system can become attracted / distracted to the details for the very reason that they can understand them and spend too much mental energy optimizing them while losing sight of a much bigger picture. Losing sight of things like the function they are spending ten hours optimizing is only executed four percent of the time and the optimization is only saving fourteen microseconds but delaying delivery of a release that will earn his company four million dollars.
Lewis’ educational background is in art and archaeology as an undergraduate and economics at a masters level -- he would not probably describe himself as highly quantitative by training or interest. His reiteration of the description of SBF’s problem solving approach mentions it can have pitfalls but doesn’t delve into its origins to convey how it can lead to deluding compartmentalization of probabilities which should not be ignored. How would you feel about flying on a plane maintained by a mechanic with this approach to missing bolts? He also misses a likely link between this fixation on quantitative modeling of “soft” problems” and effective altruism.
Effective Altruism
Lewis devotes significant time on the concept of effective altruism (EA), covering its origins and its interpretation by many involved with FTX. In a nutshell, EA is a conceptual framework by which an individual can attempt to optimize the net value they provide to society by maximizing their WEALTH by leveraging their skills and earning potential then converting the WEALTH to effort through other means to yield desired results. As an exaggerated example, a person with a unique skill worth $1 million per year in salary wanting to feed the hungry in Africa is more effective at doing so by working the million dollar gig and donating $900,000 to a charity supporting farming than by booking a flight to Chad and single-handedly trying to plant a field of wheat. Idealistically, EA is an attempt to reconcile the completing desires to “do something meaningful for others” while avoiding the need to abandon what might be very unfulfilling but extremely lucrative employment.
EA was important to the FTX story because many of the key players had been exposed to the concept in college and, psychologically, bought into it heavily. In large part due to the meltdown of FTX, the concept of EA has attracted a great deal of attention and cynicism. Besides the possibility that the FTX actors espousing EA did so in good faith, Lewis did not cover two alternative explanations, which in fact may be somewhat related.
After the crash of FTX, some observers suggested that the constant references to EA by the key players at FTX was simply a ruse, a cynical ploy used by the uber-wealthy to justify continuing to make obscene amounts of money with a promise to later mete it out to the rest of us. It was beyond the purpose of Lewis’ book to critique EA from this perspective but it IS a worthy political and economic topic. The idea that Daddy Warbucks can be excused for accumulating a trillion dollar fortune by crushing competition, manipulating markets and operating monopolies because eventually he will donate the trillion dollars building libraries, funding universities, etc. doesn’t account for the loss of opportunity created by the inequality created over the shorter term, which may trigger ripple effects that far outweigh the pot of gold forty years out.
The angle that should have been in scope for Lewis to address as part of this book was the adoption of EA as a psychological coping and rationalization mechanism. EA seems to be extremely popular with the exact demographic staffing FTX and many other financial firms. This employee demographic works in highly technical fields that utilize complex processes with hundreds of continuously changing inputs. This demographic of employees designs systems to interact with those processes to leverage infinitesimally small asymmetries to create billions in profits. All without making a dent in any real societal problem. Because the thought processes that make EA so attractive philosophically are nearly perfectly aligned with the analysis skills Wall Street finds attractive, EA becomes a easy shorthand for recruiting.
In the case of the people at FTX, the descriptions Lewis provides of how key team members lived on a day to day basis definitely seemed to rule out the possibility that they were espousing EA as a cynical means of deflect attention from the outsized wealth they were accumulating on paper prior to the collapse. Though the firms moved from California to Hong Kong to Bermuda, the actual office locations and condos they worked and lived in were more like communes inside luxury suites rather than luxury suites. Lewis states he only saw SBF in a suit twice – when preparing to lobby Mitch McConnell in DC – and even then, SBF forgot he would need socks. Otherwise, he wore cargo shorts and a tee-shirt – THE SAME OUTFIT – nearly everywhere. After the team moved to Bermuda to condos on a beach, only one person ventured out to the beach... ONCE.
From the descriptions provided by Lewis, EA for FTX proved to be both a coping mechanism for a large group of people who did not have neurotypical decision making tools for processing the dichotomy of earning great wealth and lacking any plan for its use in their personal lives. It also became a self-selecting criteria as the firm grew and attracted the attention of future employees. Those already in the firm seemed to treat the very idea of EA as a shibboleth – mention that word in an interview and you must be one of us, you’ll fit in even if I have no idea what you do or what I need – which furthered the monoculture already present from inception. The fixation on EA by the leaders of EA thus REDUCED the chance of hiring more neurotypical staff who might have had more traditional business management skills for HR, auditing, security, etc. that could have balanced the firm’s talent pool.
Neurodiverse Dysfunction
The very term neurodiverse may not be terribly effective in attempting to describe the range of behaviors and abilities as they impact work in a business setting. One can be ND in a “good way” by being higher in a desired skill without being lower in some other desired skill. One can be ND in a “bad way” by lacking a desired skill while possessing many others. Or one can be ND by being extremely high in one skill while being lower than normal at many others. That results in a large area of gray that normal language descriptions might obscure.
The narrative in Going Infinite makes it very clear that a large number of the employees at all levels of FTX were – ahem -- “very” neurodiverse. To the point of obvious dysfunction. Lewis described examples of the gifts reflected in this neurodiversity and many physical manifestations but either didn’t pick up on or chose not to spend time describing some of the dysfunction. For example, SBF hired a younger MIT classmate, Gary Wang, as the Chief Technical Officer. According to SBF, Wang designed, coded and deployed FTX’s exchange platform in a single month. Sounds monumentally gifted, right? Well, maybe. But this is the same system which was hacked multiple times for losses of $200 million, $300 million and $600 million. The same system that exhibited a glitch which began incorrectly inflating an account within an Alameda system from $8 billion to $16 billion until someone spotted it and Wang fixed it. Ooops. But Wang barely spoke a word. To ANYONE. He would arrive to work, toil for twelve hours, then leave, without saying a word to anyone. Fortunately, he seems to have found his voice after pleading guilty and testifying in court against SBF.
The funniest part of the book stemmed from the description of interactions between "the EAs" as Lewis called them. There's no need to guess what interactions within the entire group were like. The types who sit next to each other and TEXT each other rather than having a conversation or risk (GASP) eye contact. After the collapse, the nerds at FTX became a meme for the press and public alike when photos of SBF, Caroline Ellison and others came out along with letters about how the collective decided they would not subject themselves to the limitations of traditional relationships and marriage.
The outside world was just then entertaining the most lurid fantasies about Sam's inner circle. Inevitably, word had got out that the effective altruists took a principled stand against monogomy. After that, a rumor spread that the spent half their time in the Orchid penthouse finding new ways to have sex with each other. Mostly what they had done with each other was play board games. In the heat of bughouse chess matches, they'd explore every possible combination and position; otherwise, not so much. But the confusion was understandable. They'd granted themselves hunting licenses without ever really wanting to learn how to handle a gun. Who does that?
You have to read that passage twice before realizing how funny that is.
SBF was a literal poster child of the most extreme stereotype one could imagine for a poorly adjusted neurodiverse adult. Beyond the wardrobe, random bathing, mad scientist hair and manic bouncing knee, SBF routinely played video games while meeting with the press or potential investors. Bizarrely, many on the other side of these exchanges observed this, chalked it up to condescension and concluded SBF was so brilliant, he could answer their naïve questions while multi-tasking on other more important stuff. This guy’s brilliant, hand him our $150 million! (That’s nearly a direct quote from Sequoia Capital, by the way…See https://www.privateequitywire.co.uk/sequoia-capital-apologises-150m-ftx-loss/)
What should have been discerned from these encounters was that SBF was completely addled by ADHD. He was not listening to their questions or concerns. He was only waiting for key words which told him it was time to talk, at which point he would spit out whatever point he wanted to make. In the mean time, he was physically compelled to distract his brain by playing puzzles and games. These are NOT the listening skills required of any executive at a firm handling billions of dollars in transactions per day.
The interactions between neurodiverse types and presumed nuerotypicals attempting to perform due diligence is one of the most striking gaps in the book. Remember, the book STARTED because of a request from a friend for a favor to meet SBF and essentially confirm if he was "for real" and if the business was for real. Like EVERY OTHER PARTY, Michael Lewis met SBF, had a longer period to talk with him than most, and became enamored with what he wanted to see in SBF with all of his grandiose ideas about effective altruism. Lewis, like every other investor in the companies, completely ignored numerous red flags about the BREADTH of business experience any of these players possessed. And despite literally being there the day it all ended and SBF was arrested, Lewis never came back to address his own due diligence failure on behalf of his friend. Maybe the friend wound up ignoring the thumbs up from Lewis and kept his money.
Even if people met SBF and his team and were somehow wowed by the effective altruism distraction, a very brief list of questions and demands could have burst that bubble of fantasy and led to a more appropriate conclusion:
- What are the names and resumes of the people on your board?
- Provide the name and resume of your CFO and let me talk with your CFO.
- Provide the name and contact of your outside auditor.
- Provide the name and resume of the person in charge of your network and data center security.
- Provide a complete addresses of every owned, leased or rented location used for offices or data center operations in all countries.
- Provide an overview of your trading system at a block diagram functional level and a physical level, identifying all locations where customer data is stored, public portal traffic is processed and your disaster recovery plan.
An incomplete or missing response to any of those questions should have been a red flag. An incomplete or missing response to three or more should have been a hell no.
As a final anecdote that conveys the emotional maturity of the man behind the entire enterprise, there is this. On December 12, 2022 as Bermuda police came to arrest SBF, there were two remaining FTX employees hanging around, attempting to help Sam and do their own searching of files to figure out what happened and how they might have been involved. As the police were cuffing SBF, the other FTX employees went to the condo room he was using as a bedroom to try to collect anything he might find of use so he could take it with him. While sorting through the room's contents, they found a keepsake box. Opening the box, they found a stuffed animal. The stuffed animal was named Manfred and had been kept by SBF since infancy. Lewis doesn't state it but essentially, Manfred is the Rosebud of the SBF story, the sole object in SBF's entire life to which he has maintained any attachment. Does this sound like a remotely well-adjusted person who should be operating a firm with billions of dollars of Other People’s Money?
Missing / Conflicting Analysis
The narrative in Going Infinite has multiple instances where Lewis seems to round up a cohesive set of facts then fails to link them together with an appropriate conclusion or makes references across the narrative that seem to conflict with each other – either in supporting a claim made by SBF or a conclusion reached by Lewis as author.
The book creates confusion about the cash flow and profitability of the Alameda and FTX operations over their lifetime. Alameda came first as a vehicle for making hedged trades across different cryptocurrencies across different exchanges. After an initial team panic triggered by a software glitch that "misplaced" $4 million dollars of coins from a trade only to appear weeks later, Alameda began executing millions of trades per day and racking up profits. Alameda subsequently supplied roughly $10 million dollars of cash to cover the start-up costs for FTX as a second entity. Yet later in the narrative, Lewis describes different meetings SBF held to solicit investors for capital to build FTX. But the company only had roughly fifty employees at that time. And later comments from the person who became employee #50 indicate Alameda / FTX was not necessarily paying competitive salaries. #50 told Lewis their FTX salary was eighty percent less than their prior Facebook salary. It was ninety five percent lower than another offer they declined from TikTok.
So was Alameda making profits or not? And why did FTX need so much capital up front with such a small payroll? At its peak, Alameda / FTX combined only employed about 400 people. Even if they were all pulling $250,000 salaries, that would be $100 million. SBF claimed Alameda earned $1 billion in 2021. Later documents leaked after the collapse showed tax records stating $388 million in net income for 2021. And they clearly weren’t paying everyone $250,000 / year. Lewis never steps back from these facts to reflect on the larger pattern of inattention to basic financial solvency between the two firms.
More curiously, Lewis collected notes from his discussions with FTX actors prior to the collapse, compiled his own back of the envelope estimate of the combined balance sheet of Alameda Research and FTX from inception to collapse and reached a very surprising conclusion… It’s possible that NO MONEY WAS EVER MISSING. This conclusion might be driven by large loans SBF took from the companies as future donations to charities or investments in other firms which in fact never saw fruition but never had accounting records corrected, making the location and status of the funds ambiguous or completely unknown. By Lewis' reckoning, he came within roughly $100 million dollars of finding "balance" when the original bankruptcy was triggered when the firm was thought to owe $8 billion to creditors. With another group of crypto assets held by FTX that John Jay and the bankruptcy court thought were near worthless actually priced at still-current market rates, the books added up nearly $1 billion in the black. Of course there are other facts scattered through the book that Lewis seemed to omit from his rough calculations, including another electronic theft of assets of $450 million that occurred during the final meltdown. Given the fact there were several $x00 million dollar "disappearances" from the exchange – at best due to bugs which Gary Wang fixed rather than inside theft – and given the poor accounting controls, it seems far more likely FTX was in fact insolvent by several billion.
Michael Lewis as Bob Woodward
One concern that has been expressed more frequently over the years regarding famed reporter Bob Woodward is that he has slowly lapsed into becoming a stenographer of his subjects. His bona fides as half of the duo that cracked Watergate created a dynamic where famous or infamous people approach HIM to write their story. Woodward agrees, takes audio notes on EVERYTHING, uses his fifty years of experience to second guess the material dumped in his lap but pretty much restricts his narrative to what can be filtered from that material. If a topic merits doing ADDITIONAL research independent of what the subject says, it doesn’t appear in the final draft.
After reading Going Infinite, I can’t help but ask if Michael Lewis is subconsciously adopting this same laid back approach to his material. Here are a few examples. The book describes how SBF hired his friend Gary Wang away from Google and how Wang coded what became their core trading system in a single month, entirely by himself. From that feat and statements of those in the company, Lewis concludes and relays to the reader that Wang was a coding savant. Well, was he? The system he coded experienced multiple glitches over time which resulted in transactions worth hundreds of millions of dollars to suddenly go missing – for weeks or months – before reappearing. It resulted in outside hackers stealing hundreds of millions of dollars worth of tokens from accounts which were never recovered. So how good was he? Lewis never consulted outside experts to provide an alternate perspective about the integrity of the system from a logging, audit or security perspective. He never references any conversations with end-users of the platform who might have been able to provide insight on its ease of use, reliability, quirks and outright bugs as well.
In the same vein, Lewis relays to the reader accounts which imply that SBF and other key personnel in the two companies were not very adept coders. The narrative seems to imply that the extent of computer skills for most of the key players was limited to using computers for email, video conference calls, word processing and games. It isn’t important to Lewis’ story to know exactly who coded each line of the system but it is crucial to understand how development and bug fix work was assigned and how changes were tracked and who had permissions to update the code. The software of the FTX trading system essentially constitutes a crime scene, yet the details conveyed by Lewis essentially imply “silent Gary” was the only person that altered the system. It's possible, certainly with the types of players involved, but still not probable and merited more follow-through.
This lack of follow-through is particularly disconcerting because this very concept is addressed explicitly at the end of the book as Lewis talks with the interim CEO tasked with chasing down the missing billions. Lewis asks John Ray why he generally avoids talking with former leaders of the firms after he takes over. Ray’s rationale is that former players often have an axe to grind or a butt to protect (their own) so hints they provide are likely slanted. Ray believes it’s better to look at the firm’s carcass with unbiased eyes, even if those eyes lack some institutional knowledge about internal processes, politics, etc.
Lewis makes a point of identifying cases where Ray’s modus operandi results in the bankruptcy chasing a pointless lead but at the same time, Lewis seems comfortable using all of the comments provided by those same players to him during 2022 without running them by outsiders who might have a fresh perspective. More importantly, by the end of 2022, Lewis’ dealings with virtually everyone at FTX should have made it clear to him that all of them exhibited certain deficits of perception and expertise regarding basic business processes, accounting and even ethics, rendering many of their volunteered insights suspect.
The Ultimate Miss
The most surprising aspect of Going Infinite is the author’s failure to correlate his observations regarding the FTX crew of "neurodiverse", AE-fixated players with his observations from Liar’s Poker in 1989 and connect them as an inevitable progression. Lewis was there in 1984 as Wall Street first learned how to screen for a very specific combination of math skills, risk-taking behavior and aggression to institutionalize a new style of financial trading that was very successful at creating profits, if not always anything else of long term value for clients or the larger world.
Thirty four years after documenting that experience, Lewis saw a nearly identical dynamic trigger a very similar financial implosion. In the new case, the behavior selected eliminated the bro culture vibe but replaced it with a higher level of internal quantitative acuity which displaced other crucial areas of expertise. More importantly, this new mix of selection criteria was made easier to spot by having a near-perfect correlation to people fixated on "effective altruism". Moreover, this common blind spot was propagated through most of the company because the founder, SBF, found it easy to use interest in EA as an instant shibboleth during hiring and he did virtually all the leadership hiring. He wound up hiring an entire team who all had the same dysfunctions as himself. At best, the resulting monoculture was virtually guaranteed to miss warning signs that could have triggered course corrections. At worst, the resulting monoculture could have been exploited to trigger even greater losses had even one employee exercised more proactive criminal intent.
At its core, the model for operating FTX as a business is a curiously apt microcosm of the dangers of widescale adoption of artificial intelligence. FTX was founded by a person with greater than average gifts in selected disciplines amenable to computerization. That founder not only used computer modeling optimization concepts in organizing his own life, he used to them to select an entire team, resulting in a firm which built a complex system optimized for the requirements posed by its creators. But that list of requirements was woefully incomplete because the creators all exhibited nearly identical blind spots in their understanding of the complete problem. They stuck with what they thought they knew, bet heavily on it, enjoyed initial success that encouraged expansion then encountered inputs they were unable to imagine and their system failed. AI systems can and will exhibit the exact same failure mode. Any AI model attempting to process more inputs than a human can MAY synthesize additional "knowledge" to feed back into its "decisions" but such intelligence may still reflect the blind spots of the first humans who influenced its zero-generation incarnation and will still likely result in failures.
Going Infinite never makes these points directly in its narrative. However, it does provide a smoothly flowing narrative of a fascinating (in the train wreck sense) business and management failure that help make these larger realities easier to identify and apply in other contexts. For that, Going Infinite is worth a read.
WTH
