How AI *should* work; How airmiles *really* work and; Why meetings *are* the work - LuLd #009
Looking up, Looking down is a regular collection of clues and hunches for innovation and strategy practitioners
Looking up, Looking down is our regular(ish) drop of recommendations for reading, watching, listening and doing. Think of them as clues and hunches that we’ve bumped into - they’re not yet joined up into a fully-formed narrative but they hint at a bigger story and feel useful and interesting enough to share. Subscribers get them delivered to their inbox every two weeks.
Hate scrolling? Here’s the list:
Deep Learning Is Hitting a Wall - Nautilus
TOBY: Ok, this is 28’ read, I hear you! But wait, actually there are effectively two articles, so you can read the first part, and get a sobering alternative view on the state of AI and what its role is really likely to be without doing the full 28’! And who said great insight needed to be a soundbite?
I think there is wise counsel in not assuming the tech wet dream of general intelligence is just around the corner (see Gartner’s hype cycle) , as it always seems to be. This is not to denigrate the enormously impressive work AI can do, but rather to position it more in a support tool, when, as is characterised in the article, we just need a rough and ready answer.
The second part of the article then proceeds to unpick the flaws in the current model of ChatGPT and the like, which is really a look at how deep learning is just based on massive data accumulation vs symbol manipulation (basically what most software is - think algebra). It turns out the latter is exactly what you need for novelty, whereas deep learning works within a confined set of known and repeatable rules and just muscles it through with huge amounts of data. And some hybrid might be the stronger way forward.
What it also highlights is that this split , in what we assume is a classic deeply rational scientific world, comes about through some bad blood. Imagine that, our flagship latest version of AI flawed due to an all too human bruised ego. Looks like General Intelligence will include mood swings..
TOM: This is a pretty geeky deepdive into a simmering battle over how we should to ‘do’ AI, and into the ‘why’ of the limitations of GPT4 and its kin. As Reid Hoffman, one of the original investors in Open.ai put it in a book that he (obviously) wrote with GPT4: ‘A powerful Language Learning Model is a complex statistical system’. In other words, it really has no idea what it is talking about, but it’s damned good at talking. This article then, is a refreshing take from an author who is a professor of psychology and neural science at NYU - pair it with Ezra Kleins excellent interview with him and with this great meditation on the nature of sport and the controversy of robot umpires, and whether statistical accuracy or narrative drama is what we really want from our lives.
Here’s Why Airline Loyalty Programs are Profit Machines
TOM: Ah, how I (don’t) miss the days of flying to Hong Kong for a 2 hour meeting, consoling myself that at least I’d bagged an epic number of air miles to sustain my slow climb up to coveted silver status and all the epicurean delights of the British Airways lounge that it promised.
If, like me, you’ve always felt that airmiles programs were more like addiction than loyalty, then this piece is a treat - exposing the workings of the airline business model for what it is - a completely different proposition to what you might expect: in which the flights themselves are just a necessary operating requirement to back up the real business: that of a quasi-bank that prints its own money called ‘airmiles’ and then flogs it to credit card companies and car rental firms. Amazing, a testament to the creativity of capitalism and the credulity of all of us who ever kid ourselves that we’ll one day land a decent seat with miles.
TOBY: I rather like what the airlines have done. . This isn’t a consumer swindle, it feels like a weird “win-win-win” situation. The airlines clearly get value from this, not least because they are able to get funded by banks. The banks get value in terms of reduced CAC and enhanced profitability, otherwise why would they pay the airlines for it. And the consumer gets cheaper seats. Gotta say I have definitely got some good seats and enjoyed some fine lounge food off the back of this!
I don’t think the idea that airmiles are some kind of charitable act for frequent flyers is in anyone’s minds, but the fact that consumers do benefit feels like, as Tom mentions, great capitalism! And I like how the different players have analysed their own value from it. For the banks this must be a really good ROI route to acquire customers, and it would be really interesting to see the inside discussions as they started to explore that idea - not obvious that they wound brand and pay for a shared credit card.
It is like Tescos giving cash back at the tills way back in the 1990s, customers loved it as a way of getting cash without going to Cash machines, but the real benefit was to Tescos as it significantly reduced the amount of cash they had at the end of the day, and the costs of collecting and handling it.
Meetings *are* the work.
TOBY: A thoughtful look at a problem that is the bane of most people’s lives - meetings. The writer makes a strong case for what the role of meetings really should be. She has a great phrase “ success in knowledge isn’t about the facts we know but how good we are at judging the truth of uncertain things.
I love this because it goes to the heart of what meetings should be - how we use private and shared knowledge to make better decisions collectively - vs peacocking around things you do know (or more often believe you know but don’t , have a look at the “circle of competence”).
She also brings to life the challenge of businesses living in operational efficiency, where everything is known and planned. Knowledge work is necessarily messy and unstructured, it needs space to explore and learn. Businesses that try to rationalise and optimise the crap out of every minute are completely absent a way of figuring out the future. This creates bad meetings, where the purpose is to manage, control and transmit, not to explore and get shared learning.
Meeting is work, and often the best kind of work. Tom sort of alluded to this in a previous Looking Up Looking Down when he commented on something I posted about a low meeting culture. Between that article and this I can see what sort of meetings we should be having and why.
TOM: It’s funny how meetings can be either tremendously energising or utterly soul-sapping, perhaps it’s because meetings have become ‘thingified’ - we see them as events, not actions. Great meetings are about exchange, exploration and action, they can be the clarifying moments in projects, initiatives or businesses where progress is made. Sadly, as Toby says, they too often become about presence and peacocking. I particularly like Amazon’s principle of Bias for Action, it’s a great counterpoint to a world where we have too much data, too many tools, too many options for procrastination in the name of ‘getting certainty’. In my experience, great meetings and great meeting habits can lead to great culture, but starts with understanding what they are and what they’re really for.
Why didn't DeepMind build GPT3?
TOM: This is not an article about AI. It’s much more interesting piece on the cultural differences between two competing firms. Remember when DeepMind created AlphaGo, the AI that taught itself to play Go and then beat the world’s best player (a feat thought impossible at the time)? This piece, written by a former DeepMind employee looks at why GPT3 came out of a different organisation, with some interesting observations around how companies create success measures, and the knock-on effects of valuing academic prestige vs valuing engineering attainment. I think this is also a great example of the different risk profiles of start-ups vs scale-ups. The risk to DeepMind (now part of Alphabet) of shipping a speculative product early is so much greater than to Open.ai - what will be interesting to see is how long Open.ai’s culture sustains, as success, valuation and jeopardy climbs.
TOBY: it’s an interesting article, that at its heart seems to be saying that because we aren’t geeky academics trying to get peer reviewed to fame, we get shit done. Which really is a statement about internal culture and metrics, which I totally buy into. I see that A LOT in engineering companies where value is disproportionately in smartness of solution vs does anyone want it (more commercial lens).
I am mixed on the self congratulation of not sticking to the targets/ absence of any metric. On the one hand, sure, if the metric and incentives turn out to create weird outcomes that become really obviously wrong then ditch and move on. On the other hand to just declare victory with whatever, feels like the increasing tendency of too many start ups to use whatever formula (vs GAAP) to maintain that they are now profitable (if they exclude whatever they feel like that drags on costs).
AI Looks Like a Bubble
TOBY: In a related sort of vein, a really interesting article about the AI bubble, about how technology innovation does not equal investment opportunities (its the business model dummy), supported by a great deep dive into the only B2B AI software company and how it demonstrates this is so definitely a bubble.
The article really brings to life the difference between a technology (and he does a good breakdown of how to think about the AI value chain) and being able to apply it for commercial benefit , and the need for a clear understanding of both the likely ecosystem and a line of thought to how revenue is generated. A great reminder of not getting bowled over by the latest buzzy tech craze, but rather deep dive into the complex working of how that technology is expected to make a difference and what that actually implies. This is part business model part ecosystem.
TOM: Worth reading if you’re interested in what it might take for AI to stick as a product and business model. Certainly beautifully summed up by this line: “A market is still subject to market dynamics, regardless of the level of science involved.” This is also a great dissection of a business in the spirit of narrative and numbers.
And finally this: “the hard part isn’t the AI. It’s doing the change management” and that, my friends is the essence of ten thousand feet and two inches right there. From 10,000 ft AI looks like a complete slam dunk, and it no doubt will be in some form, but down at 2 inches, where you’re trying to figure out what to exploit and how to exploit it - whether you’re selling it, or trying to profit from it - is where shit gets real, where human systems, incentives and inertia can be a horrible reality check. We haven’t yet seen the right shape of AI-driven businesses, but we may be starting to see the wrong ones.
In For a Pound
A nicely written article about what what you *actually* get when you purchase one of the Royal Opera House’s £1 tickets. This feels particularly apt in the month in which Silicon Valley Bank was sold in the UK for just a pound. One reading of this is as a rumination on price and cheapness, but really it makes me think about value propositions, and where the value lies. At the heart of any business there has to be a value proposition that matches a ‘thing’ that customers want or need, with an ability for a business to deliver that at a price which is desirable to customers and viable for the business. Are £1 tickets really good value? It probably comes down to why customers are really buying them. Or even, in these time of constricted funding for the arts, whether we understand who the real customer is: maybe the real value proposition is in the signal that these schemes send to funders and benefactors…
TOBY: What is interesting is that the article talks about value when in reality it is about price like Oscar Wilde’s cynic - the man who knows the price of everything but the value of nothing. The value of say a McDonalds meal deal is steady (appeal, nutritional value, etc), even if the price goes up. It is good or bad value in relation to what you can buy for that sum and how much of your income it represents. . A pound for an opera ticket in 1800 would have been the equivalent of £107 in today’s time, and the average income was around £25.50. Would you spend two weeks wages going to the opera?
The author’s framing (and belief on good value) is timeless in an economy that is anything but timeless. What I take from this is how people (a person) anchor on a sense of value at a certain price and struggles to let go in a sort of “in my day you could get a pint of beer for 10 shillings” sort of logic. That is an interesting dilemma for businesses, in the lag between rising cost of goods and consumer perceptions of value - probably like right now with 10.4% inflation an the fact that my Greggs Sausage roll (in central London, they have lower prices as you move out) rose from £1.20 last Christmas to £1.45 today. The fact that I still buy means I still attribute value.
On the author’s “is £1 good value”, it is such a personal question it is unanswerable to have a general view point. For a business it becomes about what happens to the buyer. A bit like the airmiles argument above, if I typically have empty seats at the opera (fixed costs) on a Wednesday night then selling them for £1 has marginal contribution so why not? It makes perfect business sense.
If I can also create an ease of entry into the opera world and get some occasional goers to turn up (I’d go and see “the Barber of Seville”) then maybe I have just recalibrated their sense of value so they would pay £107 for a ticket. Business wise it makes a lot of sense. The price isn’t the value, and even the thing you buy isn’t necessarily the thing the business really wants to sell, maybe they are selling you your gateway drug to the world of opera….