Predicting the Potential of “New Electricity” by Looking at the Lessons of “Old Electricity”

Andrew Ng (former Chief Scientist at Baidu, Co-founder of Coursera) is fond of referring to AI as the “new electricity”. Similar to how electrification ended up transforming every single industry during the 20th Century, AI is broadly expected to have similar (if not greater) impact in the coming years.

This post by Oscar Li does a great job encapsulating Andrew Ng’s thoughts as shared during a recent lecture at Stanford.

The social and business ramifications are easy to extrapolate / imagine: Anything that you can do now, you can do better with more AI and data. And as the technology that underpins AI continues to improve and the datasets continue to grow, AI will allow us (or machines) to do whole new things not possible before.

If you’re like me, it’s actually quite overwhelming to ponder the full implications of AI. But if AI is on track to transform the world like how electricity did, we should be able to learn some lessons and create some useful predictions by looking at what happened a century ago.

Lessons from Electrification:

  1. Everyone Electrified and then Electricity was no Longer a Differentiating Factor: Today, companies that use AI and data have an advantage over those that don’t. A similar dynamic existed a century ago. Retailers that were electrified had an advantage over those that didn’t. Factories that were electrified had an advantage over those that didn’t. Eventually all businesses either adopted electrification or were out-competed. However, once electrification was fully borne out, it was no longer an advantage. A similar dynamic also played out with air conditioning in the middle of the 20th century where early adopters had an advantage, but once every retailer adopted air conditioning, it became a necessity rather than an advantage.
  2. Electricity Changed How Business was Done But Had Less of an Impact on the “Jobs to be Done”: Electrification allowed businesses to extend their hours of operations, drastically increasing efficiency and productivity. Electrification also brought about new machinery and tools that similarly improved efficiency. In other words, electricity made businesses (and people) better at their craft but had less of an impact in terms of the goals that businesses pursued. Retailers remained retailers. Manufacturers remained manufacturers. Service providers remained service providers. This is perhaps the most interesting takeaway (or contrast) with current assumptions around AI – namely that AI will necessarily lead to “new” products / services.
  3. The New Business Model Created by Electrification was the Electric Utility: Although electrification had large ramifications on how existing industries operated, the only key new business model to emerge appears to be the electric utility to provide the fuel for the disruptive technology (I’m not quite sure if this statement is entirely accurate – if any readers have supporting evidence for / against, please do share). Similarly, how many new business models should we anticipate in the coming AI explosion? It’s clear that AI will be important for existing industries, but how many new types of businesses should we expect? Perhaps a “data utility”, but what else?
  4. Other than Utilities, No One Made Money on Electricity: Once every business adopted electricity, it was no longer a differentiating factor by and of itself. As a result, electrification become purely a cost of doing business. Only utilities made money off of electricity. Can this be extrapolated to AI and data? Will it purely become a cost of doing business?
  5. Electricity Led to a “Cambrian Explosion” in Hardware: Although electrification appeared to merely change how businesses were run, it did lead to many new hardware products such as light bulbs, phonographs, radios, televisions, etc. Turned out that the best way to monetize electricity other than operating a utility was to produce hardware. This was the genesis of General Electric. Perhaps that is also what we should expect of AI.

These are just a handful of lessons. There are probably more that can be added to this list.

Taking these into account, it’s not outrageous to ponder whether AI can truly be monetized if not offered as B2B service (e.g. the utility model) or bundled with hardware (e.g. Amazon Echo, Siri / Google Assistant / Cortana in devices). Are there alternate monetization pathways if neither of these are feasible?

One thing that is an interesting contrast between AI/data and electricity is that AI algorithms and datasets are not commodities like electricity. Every utility’s product is the same. But AI algorithms and datasets are not interchangeable. What are the implications of having non-interchangeable “fuel” for one of the most disruptive technologies of the 21st Century?

What Apple, Google, and Facebook’s Business Models Tell Us About Their Ability to Adapt to the Coming AI-Future

 

Judging from market multiples, it’s clear the market is skeptical of Apple’s ability to continue to succeed and has been for many years (currently, 1-yr forward P/E of ~15x despite recent re-rating), while holding limited concern for Google and Facebook (both 1-yr forward at ~26x).

This concern is understandable given the ever-changing tech environment and the long list of tech companies that have been buried over time (e.g. Nokia, Blackberry, Motorola). Even industry stalwarts of the bygone PC-era such as HP, Dell, and even Intel/Microsoft now stand on far weaker ground. And this concern for Apple is not new as Horace Dediu points out at Asymco

However, I contend that Apple’s business model is far more robust than Google’s in the coming artificial intelligence (AI)-revolution. 

Leaving aside potential differences in technical competency, AI is likely more complementary to Apple’s business model, but potentially highly disruptive to Google’s.

And to understand why, it requires understanding the job-to-be-done for search users and how Google’s ad-centric business model fits in.

The Assumptions Behind Search

Let’s start with the user – why do Google users search? What is the job-to-be-done?Seemingly simple question with a straight-forward answer: We are curious about many things and don’t have perfect knowledge. Google can help us answer what we don’t know. For most of Google Search’s existence, the service has approached this job-to-be-done by returning a curated list of search results that are most likely to match what the user is looking for. These are organic results that are designed to improve over time as the algorithm takes into account the answers/results users found most useful in the aggregate. Fairly simple and straight-forward.

So now turning to the business model – Google Search makes money on the ads that appear alongside search results. Every click on an ad (as opposed to an organic result) generates a small bit of revenue for Google.

Under what circumstances would a user prefer an ad over an organic result? 

Philosophically, Ben Evans (A16Z partner and almost always more than a few steps ahead of the curve) hits the nail on the spot:

In essence, the existence of ads on Google Search is the most egregious effigy to the failure of the underlying search algorithm. Every time a user clicks on an ad, it is an implicit acknowledgement that Google Search did not accomplish its job-to-be-done, that the desired answer would not have been provided had the advertiser not paid for that placement.

Given a long enough time horizon, if Google Search is truly improving, barreling towards a future where Google can answer any question without doubt and with perfect context, Google’s business model would have to evolve because ads do not have an obvious place in that future. 

And that future is coming closer with the advancements in AI. 

Search and AI

What could search look like in an AI-driven world?

It might not look all that different – perhaps Google’s I’m Feeling Lucky option or the knowledge graph cards.

But, it could also be radically different such as the search results provided by assistants such as Apple’s Siri or Amazon’s Alexa.

Regardless of the ultimate direction, it’s clear that ads have a far less obvious place in this future. If AI can be used to surface the right answer the user needs, the user would have less of a need to click on an ad.

I think we’ve seen the first instance of this issue with Google’s recent Beauty and the Beast ad on Google Home. The Google Home assistant (like Alexa) does not offer a natural channel for ads, and as a result, Google will have to find increasingly creative ways to adapt their business model for the changing environment. But history is filled with examples of failed business model surgeries.

It also makes Google’s persistent pursuit of a hardware/software business model a la Apple much more understandable. Perhaps selling hardware isn’t just about protecting access to market (after all, it’s unlikely that Android will be unseated now that it is so entrenched, so why continue with the direct hardware efforts?), it could be a deliberate attempt to evolve their business model to ensure financial relevancy in the AI-driven future.

Why Investors Seem Complacent

I contend that investors (or analysts) are confusing two separate concepts to be the same thing – the universal desire to know (hence search) and the need for ads to provide that information.

There is no doubt that search queries will only grow. It’s not farfetched to imagine a future where there are 7 billion search users instead of the less than 2 billion today. But that’s a far different conclusion from assuming Google’s financial future and business model are robust as long as search queries grow.

Search is robust, but are ads robust relative to our AI-future?

Apple’s Business Model is Far More Complementary to AI

Apple’s business model, on the other hand, is much more aligned. Sell the best products available for customers’ job-to-be-done.

AI is not orthogonal to this pursuit and business model. The only question is whether Apple has the technological capabilities to do so. However, even under the assumption that Apple is not the leader in AI-technology, it is not clear that it matters. After all, Apple was not the original leader in GUIs (Xerox PARC was). Apple was not the original leader in phones (Microsoft, Palm, Blackberry, Nokia, etc. were). Despite not being the original leader, Apple had the business model and insight to ensure that they could ride the underlying technological trend, and if I were to take a bet today, I believe Apple’s business model makes them more prepared to usher in an AI-world than Google is. AI is a feature to Apple. AI is a potential danger to Google’s business model.

From a business model standpoint, Apple is far more robust in an AI-world.

What About Facebook?

Facebook’s an ad-driven business model. What about them? Surely, Facebook is potentially in danger as well.

Facebook’s situation is less clear cut because the job-to-be-done for users is different vs search. Search has a “right” set of answers. Facebook’s feeds do not. Users are not necessarily searching for anything specific but rather using the services as an outlet for time. The “wrong” answer in a search query is far more obvious than a “wrong” answer in a Facebook feed.

And if AI can be used effectively to ensure that organic content on News Feed, Instagram, etc. are more relevant and engaging, then perhaps that can even offset any deterioration in experience from an increase in ad load. AI can never make Google’s organic results more engaging to the point where a user will tolerate a less user-friendly ad environment.

Facebook’s job-to-be-done is to, cynically, offer a time sink. Google’s job-to-be-done is to get you your answer as fast as possible. One of these is more complementary for an ad-driven business model. 

Concluding Thought

Is Google as robust as investors think? Borrowing from Peter Thiel – is this something that could be true that no one believes? Is Apple in as much danger as the average person believes?

Disclosure: I have no direct beneficial interest in AAPL, GOOGL, or FB as of publishing date and have no intent to initiate a position within the next 48 hours.