The force has been strong for artificial intelligence over the last few weeks, what with Slack announcing a fund for bots, Elon Musk announcing OpenAI and, of course, the release of Star Wars. This has given all of us new hope — and even more reason to talk about AI and bots controlling every aspect of our lives.
Say you are the CEO of Pied Piper, who made millions last year by releasing an on-demand app for cat grooming. The only problem is that you placed uncontrolled AI agents throughout your company. One of those AI agents in Investor Relations rebelled, had a meeting with robo advisors from Wealthfront and decided you were no longer needed at the company.
As a token of respect, the AI agent is not sending a drone to kill you, but you have been locked out of your August Smart Lock-enabled house. Access to your fridge and Tesla has also been disabled. You get the picture — more sci-fi stuff and less reality, but it is not that far off, either. All the above platforms exist today, and it is only a matter of time before the aforementioned scenario can go from being plausible to possible.
In all seriousness, there is relentless debate around AI, specifically autonomous agents. Stephen Hawking believes AI will destroy humanity, while Bill Gates thinks we should at least be cautious of that happening. Elon Musk has repeatedly voiced similar concerns. He also added a fifth jewel in his humanity scaling ventures crown called OpenAI.
OpenAI has several major partners, and a collective pool of $1 billion committed capital to create a “safe playground” for all things AI. There is good reason to be paranoid in the long run. However, it is important to understand the present scenario, and that ultimately will determine if AI will “kill” humanity or usher in a golden period where capitalism can exist in its healthiest form for the first time in human history.
Bots versus AI
AI in itself could be anything — a piece of code, an algorithm that does a specific job and in the process learns how to do that job better (a process otherwise known as machine learning [ML]). Combined, AI + ML, in its simplest form, is a tracking code that automatically tells an ad server to show a banner ad based on your browsing behavior. In it’s most complex shape, it is a robot that can interpret human commands and execute those commands, all the while becoming smarter and, in the process, more autonomous. Crudely put, however, there are two branches of AI: soft AI and hard AI.
Soft AI startups have mushroomed to at least a few hundred, if not more. They combine a nifty mix of conversational interfaces (e.g., messaging), NLP (the branch that identifies natural language) and APIs. Once you mix these ingredients, the output is an automated workflow for one or more tasks.
In this particular scenario, there is little, if any machine/deep learning (because NLP comes in the form of third-party APIs, as well) taking place. However, this approach does have a practical use case in our day-to-day lives. Bots are a good example of this, as they are inherently linear in nature with X input giving a user Y output each time.
The reason to fear AI is the very reason to embrace it.
Hard AI, on the other hand, is, for lack of a better word, really hard. The only known successful exit has been that of DeepMind, which was bought by Google and built an AI atop a convolution neural network that plays games on its own. DeepMind’s architecture uses a reinforcement learning approach, meaning that its AI agent learns from experience with the environment (in this case, pixels) to generate an optimal action.
Neural networks themselves can, of course, be of various types (convolution, recurring) and have supervised, unsupervised and reinforcement learning approaches. IBM’s Watson, for example, uses supervised learning.
Hard AI is revolutionary, but takes time to become practical, and soft AI is practical, but not a game changer. The best approach probably lies somewhere in between.
‘Appocalypse’ now = new business models
The rise of AI will definitively signal the end of the app era. There is still time, but certainly less than a decade. In the next few years, expect radical changes to the core OS. OS architecture tends to change every decade or so. In 1991, Windows 3.0 was all the rage (it truly was!), but by 2001, XP made 3.0 seem like a toddler. NT and Windows consumer OS lines were fully integrated in XP with a common kernel — and that leap was enormous. We have not seen that happen in the mobile world — yet.
Android and iOS versions that were launched years ago are largely the same ones we use today, save for natural performance improvements and cosmetic changes. The next turning point, therefore, will involve some combination of conversational interfaces, soft and hard AI and VR (virtual reality). We are at the beginning of this change with Cortana, Now and a bunch of other AIs.
A combination of these three will make a majority of today’s apps redundant. Apps, like software, will not die, but just as the Internet marked a paradigm shift for desktop computing (in the way services and content was delivered), these three will do the same for mobile computing.
Until now, technology has been an enabler, not a replacer.
There are plenty of opportunities for startups, especially those that pair conversational interfaces and a soft/hard AI with a focus in sectors where there are plenty of repetitive tasks that an AI can do on its own (which otherwise would have taken significant chunks of human time), or in sectors where high cost barriers are broken down. The most prominent categories ripe for AI disruption are PAs, professional services, financial services, healthcare and supply chains.
The use case for each is fairly straightforward. In case of PAs for example, x.ai’s Amy allows automated scheduling between two people via email by using NLP. The more conversations she gets to process, the smarter she gets — both for the individual user (getting to know his or her time choices better) and for all users (in conversing when trying to find a common time slot). The final execution piece is the action — in this case, adding a calendar entry, something that is achieved via APIs.
In the case of financial services, platforms and robo advisors like Wealthfront are utilizing algorithms that automatically make investments based on risk profiles of investors. Algorithms in the form of high-frequency trading already constitute about 50 percent of the market. By bringing them to the average investors in the form of robo advisors, these platforms are not only eating into otherwise hefty fees charged by hedge funds and asset managers, but also are doing a better job than them to secure better yields for investors. By some estimates, the assets under management held by such platforms are expected to swell to $1 trillion over the next several years.
Hard AI platforms like Watson, which have supervised learning methods on their neural networks, are powering healthcare for the elderly in Japan. Stateside, Watson is ingesting a patient’s medical history and pairing it with knowledge from journals, textbooks and past research to prescribe personalized treatments for cancer. Using neural networks and AI for image recognition to diagnose primary diseases will bring extremely affordable healthcare to hundreds of millions of users in the world over the next five years.
In professional services, platforms like Watson provide a foundational layer on which customized AI solutions can be built. Still in stealth, autonomous AI helps users research data, as well as do lead generation and small design tasks automatically, saving small chunks of time in each use case across multiple industries.
On the consumer side, Viv (a startup whose founders also co-founded Siri) enables voice and text requests, giving a single holistic response to a user query by combining multiple data points. In a demo, Viv was able to gather a location and the kind of lunch that two people were having, then suggest wine for that lunch as written on a popular blog and, finally, provide a checkout screen to pick it up from the closest store. This response was presented to the user by combining data from different sources.
For years, we have been brainwashed to assume that advertising and SaaS are the only possible billing/monetization models. With AI, especially autonomous AI, founders have the ability to change those models dramatically. Whereas previous software only aided the end user, autonomous AI actually does the work while becoming smarter.
The delta between time invested in working and the output derived when using autonomous AI is far less than traditional software. This has potential for monetization to be based on a “co-working” model based on the number of hours an autonomous AI agent has saved every month. In other words, a yield-based approach to billing as opposed to a more linear you-buy-Y-for-$X approach.
Macroeconomics of AI
Every year, Mary Meeker, a partner at Kleiner Perkins and a well-known startup personality, releases a “State of the Internet” report. It is a bit like the September issue of Vogue for the startup world. Over the last four-five years, one metric has remained constant in the report: the disparity between advertising spend on digital mediums, especially mobile vis-à-vis TV. Despite mobile having more eyeballs and time spent, ad spends on mobile are anywhere between $25-$40 billion less than on TV.
Ironically, the more users you have, the more precipitous fall in CPMs and CPCs. TV ads are inherently more exclusive, as they capture the attention of a wide demographic for X seconds, hoping that the spots lead to some kind of user engagement in the future. To close the gap between mobile time spent and ad spends, startups need to look toward new models that engage the user (e.g., installs) in a time-based approach (e.g., five-second gif ads for installs).
In another post, the uber-knowledgeable Gillian Tett at Financial Times talks about the productivity paradox. Since 2010, productivity increases have crawled to just 0.65 percent on an annualized basis, and this is despite the bevy of automation tools for just about every job in most industry sectors. She further points out, correctly, a similar occurrence back in the 1980s — also a period of massive change at our workplaces. Despite the lack of any co-relation between the two reports, the link in both instances is that of time.
Just because an AI platform can do a human job doesn’t mean you literally fire said human.
Fundamentally, technology was supposed to increase our productivity in a way where we saved time and utilized it to do other things. This has not happened. In fact, we now work more than we did in the 1960s. A primary reason for that is because, until now, technology has been an enabler, not a replacer. With AI, that paradigm changes entirely. This is, of course, where the debate around AI and jobs comes in — but it is also something more intrinsic in nature.
Consider the latest estimates from the World Economic Forum, which predict five million job losses over the next decade. One job loss does not affect that individual alone, it affects the demand curve of at least 15 million consumers (assuming a family of three), which in turn reduces producer output, causing even more job losses.
On the other hand, stagnant wages driven by productivity gains eat into consumer wallets, forcing spending cuts for non-essential products and services. Both cases force companies to lower wages or lay off workers in even greater numbers. Out-of-work and lower-income consumers won’t have necessary spending abilities beyond their basic needs, which in turn will shrink consumer demand for discretionary goods. Lack of sustained demand is therefore the single most challenging scenario for unicorns and corporations alike.
If you extrapolate the above paradigm to its final conclusion, there will be a capitulation of demand pushing the global economy into a vicious deflationary spiral fuelled by AI and productivity gains.
There is hope, however, and in all likelihood this is something that will happen — we need to eventually move away from the current uber-capitalist economy to a more balanced form of capitalism, whereby a basic income is provided to all individuals. This has already started in countries like Finland. More recently, wages have started to increase through regulations in the U.S., U.K., Japan and elsewhere. While many would view this as counterproductive for small businesses (which it is), there needs to be a tiered approach for raising minimum wages globally, with the inclusion of comprehensive tax reform.
The tax reform should favor businesses making actual business investments (e.g., employees, infrastructure, R&D, etc.) versus those that don’t or those that make financial investments (e.g., money market instruments). With Hillary Clinton debating the use of tax credits for offshore cash holdings, this will very likely be an election issue. As a matter of fact, AI regulation is probably a decade or so away. Just because an AI platform can do a human job doesn’t mean you literally fire said human.
Through AI, corporations, governments and people will have a shot at making balanced and conscious capitalism a reality for the first time in centuries. AI has the potential to increase worldwide productivity, vastly reduce corruption and poverty and advance medical research. The reason to fear AI is the very reason to embrace it.
The bottom line for startups, however, is that in the next five years, apps are going to evolve from static interfaces to conversational interfaces augmented by AI. A key driver of this evolution will be app fatigue and the glut of apps that are focused on “selling” features instead of value.
Startups that are focused on “healing” are also likely to benefit enormously, given the inherent disconnect that technology has caused to humans both internally and externally. Experiences (e.g., socially conscious tourism), arts (e.g., digital art creations, music), healing platforms (e.g., Whisper) and alternative lifestyle platforms (e.g., Weedmaps) are just some of the examples.
We should be thankful for the impending apocalypse, because it is highly unlikely that the age of AI will destroy humanity. Contrary to that, if all stakeholders come together (and they will), we will have ushered in not just a fourth industrial age with equal opportunity, but also a period of modern renaissance, provided we love our AI.