Translation by Elad Gil: Translation by Xue ‘er Xu

Although tesla’s recent accident with its driverless car has cast a bit of doubt on the technology, it’s still a direction that many big tech companies are keen to pursue. Ai and machine learning have moved beyond the conceptual assumptions of the movies, from wearable devices that monitor health data to intelligent robots that are actually happening in the real world. Perhaps the most controversial aspect of machine development is the substitution and morality of mass labor. In this post, Elad Gil, Twitter’s director of strategy and a noted investor, predicts what he sees as possible growth around machine learning and ARTIFICIAL intelligence, as well as the impact of the real AI era on society.

In the foreseeable future, AI technology will play an incalculable role in many business fields. Over the next 10-20 years, machine learning may gradually replace white and blue collar jobs, eventually leading to massive social upheaval.

So far, most mainstream coverage has focused on the general uses of AI (so-called AGI strong AI, which refers to ai that is fully capable of performing human mental tasks), but little attention has been paid to the specific vertical markets where AI and machine learning are most likely to transform over the next five years. In short, I think we’re still more than a decade away from truly strong AI, but ai-powered vertical products are going to be a big change in the next few years.

Here’s a list of areas where entrepreneurs and investors are undervalued, and where I believe a number of large AI companies will soon be growing. In all the fields around machine learning [1], there are two key points for a startup to succeed: 1. The ability to build a useful database for training models, and to have a simulation environment for recursive testing and closed-loop feedback on the models; 2. Choose a market and let machine learning create products according to market needs. Making AI products for the sake of making AI is like failing to find the answer to a problem, and usually such startups are prone to failure [2].

Here are the areas where I think big AI companies will emerge in the next five years:

1. Hardware and integrated circuits

The inexorable momentum of driverless cars and other markets that make heavy use of machine learning will drive demand for more efficient hardware optimised for machine learning models. Few startups or investors have invested in the development of a chip architecture that is the foundation of a faster machine learning system. Many companies that use specialized clusters for machine learning use NVIDIA gpus, which are not specifically optimized to be the best choice for machine learning models. So there is still a lot of room for innovation in this area of hardware (ASIC or otherwise), and there may be big companies like ARM or Qualcomm. This is driven not only by the spread of machine learning across industries, but also by the massive demand for processors in driverless cars and other hardware. Perhaps the first billion-dollar company in artificial intelligence will be in chips. [3]

Companies such as Cerebras[4] and Nervana, which specializes in deep learning, are currently working in this area.

fintech

The rise of WealthFront and FutureAdvisor (acquired by BlackRock) in the “robo-advisors” space suggests financial services companies are starting to focus on machine-driven portfolio management and trading models. Machine learning is having an increasingly significant impact, whether on portfolio building and trading, or extracting analytical results from different types of financial data.

The use of machine learning and big data in fintech will go in at least three directions:

  • A. Tools to improve trading efficiency or collect unique analysis. OmegaPoint, for example, is focused on building a “new Generation of Bloomberg”, building machine learning models on data services for trading.

  • B. Portfolio management and trading operations based on machine learning.

  • C. Use machine learning models to understand financial products and price them reasonably.

Products like insurance, mortgages and other derivatives will benefit greatly from the application of ARTIFICIAL intelligence. If a startup or current player in the industry can use new statistical methods and machine learning to figure out better mortgage installment plans for customers, it will be a big company.

3. Autonomous cars and trucks

The emergence of driverless cars will disrupt the multibillion-dollar transportation market. Car and truck makers are aware that an existential crisis is brewing. The emergence of Tesla, which now seems to finally have a car for the masses and is making a big push into driverless cars, has added to the crisis. To make matters worse, tech giants like Google and Baidu are also getting involved in driverless technology. Many car companies prefer to grow on their own and avoid relying on tech firms — hence General Motors’ $1 billion acquisition of Cruise, a start-up, and Uber’s 1% stake in Otto, a self-driving truck company. In the next one to three years, mergers and acquisitions in the autonomous vehicle sector will continue to be frequent.

The development of driverless technology will inevitably lead to millions of job losses and significant social change (much of it economic deflation). Although the industrial revolution happened 150 years ago, it is not surprising that there will be a bigger wave of unemployment over the next 10-20 years as many jobs disappear and people are replaced. Political unrest is inevitable if the government does not provide the unemployed with alternative livelihoods.

4. Drug

When people talk about artificial intelligence or robots, they will say that blue-collar workers will lose their jobs, but I think most white-collar workers will be replaced by artificial intelligence. [5]

There’s going to be a big shake-up in medicine. From disease definition and diagnosis to treatment, machine learning will transform the entire medical system. VinodKhosla, the king of “technology” investments, has plenty of ideas.

In medical care, machine learning also has great potential to replace much of the old architecture. Imagine a future where everyone with a smartphone has the equivalent of the best doctor in the world, at low cost and on call.

A. Diagnosis Remember that awful trip to the doctor? You wait in line for 45 minutes, and then you watch it for five minutes and then give me some advice. Poor customer service and high medical costs are no way to survive in any competitive industry.

The shift from doctors to machines will boost individual industries for each disease — IBM’s Watson cognitive computing project, for example, has announced its own advances in oncology. Similarly, there have been startling advances in the diagnosis of depression and the study of other mental disorders through deep learning on computers. Machine diagnostics are often limited by the amount of data available and the availability of closed-loop feedback on diagnostics and results.

To accelerate the use of machine learning in medicine, one option is to buy an existing radiotherapy center or clinic. The radiotherapy center will be connected to the Internet to optimize data generation, which will be used to train machine learning models to diagnose and treat patients. By implementing machine models in clinics that go hand in hand with traditional standard clinic services, you can reduce regulatory and patient care issues while getting closed-loop feedback from the machine.

At the same time, the implementation of machine learning models will improve the accuracy of diagnostic tests. For example, using machine learning models to determine whether variants were correctly identified on DNA sequencers or which cell types were observed on fluorescent-activated cell separators. My company, Color Genomics, has started to apply machine learning to genetics in a different way.

B. Treatment and diagnosis are similar, and machine learning will help patients choose the right treatment. And perhaps the biggest limitation is access to data.

C. Continuous monitoring and analysis One way to add data available to machine learning models is to adopt new consumer-led technologies for continuous monitoring. Self-regulation of health data has a small group of loyal supporters. [6] Companies like Cardiogram are giving consumers more control through continuous pulse checks and other data monitoring. A number of Silicon Valley residents use The FreestyleLibre glucose monitors to monitor their own blood sugar levels. Promoting the development of “science for all” and arousing people’s awareness of taking an active interest in their own health care and well-being could change existing medical practice.

5. Education

The US education market is pretty bad from a tech perspective, so I’ve never invested in any education startups. Even so, I have a somewhat pessimistic expectation that companies will emerge that develop smart education systems that tailor what students learn online to what they learn offline. This educational technology could dramatically improve the learning abilities of students in developed and third world countries.

6. Other areas

There are many other areas (manufacturing, advertising, etc.) that have been or will be disrupted by machine learning that are not covered in this article. As an entrepreneur and investor, I am personally most interested in these areas and the many growth opportunities around them. Instead of focusing on the value of machine learning itself, entrepreneurs should think about how machine learning can make a product in a market ten times stronger. And that’s the key to building a big company around AI.

Data limit

Fundamentally, much of the AI industry’s growth is limited by its reliance on data volumes. Many advanced machine learning applications may survive if data in finance, healthcare and other fields are harnessed in large quantities. Large companies (Google, IBM, etc.) and startups can build useful databases in one of two ways: by buying access to data or collaborating on solutions. Data will transform many industries, be a force for the cheapening and democratisation of technology (for example, standards of health care are becoming more uniform between rich and poor countries), but it will also displace much of the Labour force in rich countries. Over the next five to 10 years, machine learning’s ultimate impact will be in broadening access to critical information (such as medical diagnostics) and displacing millions of workers. More to come.

This article originally appeared in Elad Blog, written by Elad Gil, translated by Xue ‘er Xu, Translated by ONES Piece. ONES Piece is a non-profit translation initiative launched by ONES Ventures that focuses on technology, venture capital and business.

[1] Of course, once we get to truly strong AI, the world will move at a rapid pace, but that’s still a long way off, and the exact timeline is nothing more than speculation. ↩

[2] Or acquired by Google, Facebook, Uber, etc. There will be thousands of acquisitions in this market as all companies sharpen their capabilities in this area. ↩

[3] You can really argue that Google Search/Ads is the most successful vertical application of machine learning in the world, with a market value and revenue of over $10 billion. So I mean, the next startup that will succeed :). ↩

[4] Special thanks to Andrew Feldman for reading this post. ↩

[5]: There will be more discussion on this topic in future articles. ↩

[6] Many world-changing products may at first seem like childish toys, and self-monitoring products with health data may follow this trend. ↩