Shulou(Shulou.com)06/02 Report--

The domain name has more than one AI, and the financing may be 50% more. For the sake of "dream", it is the best policy to draw a cake first.

Tidying up & Writing | Yu Duotian

The phrase "fake it till they make it" (pretend to do it until you do it) has been very popular in entrepreneurial circles.

Given that many startups and mature companies don't release their products until after 100% success, this seems to be a default operation in the circle. But there is one question:

When it comes to developing artificial intelligence technology, how much is too much when startups are also fake it till they make it?

If the bow is full, it will break.

The price of access to capital and buffer time is the huge risk that "white lies" will be exposed halfway before they are converted into reality.

Last week, the New York Times revealed that the AI disaster response solution produced by star company One Concern did not live up to its name, and some of the results of disaster prediction were considered by post-disaster experts and engineers to be fatal defects.

Today, the Wall Street Journal made another disclosure:

Engineer.AI, an Indian startup that claims to be building an artificial intelligence app development platform, does not really use artificial intelligence to develop applications.

On the contrary, the real contributors are the staff who use manual methods.

According to public speeches and promotional materials by Engineer.ai founder Sachin Dev Duggal, an artificial intelligence software called Natasha developed by Engineer.ai can help anyone create a customized app.

In other words, anyone can quickly create a mobile app by clicking on the menu on the site with the help of this AI aid. The general process is as follows:

Users can choose any existing application template they like (for example, the example given on the company's website is app for ordering pizza). Natasha then creates a similar application to a large extent automatically.

The company says that because most of the code that supports popular applications is similar, its "artificial intelligence software" has mastered most of the structure to help users automatically assemble new applications.

This will make the whole process cheaper and faster than traditional application development.

As for the effect, the Wall Street Journal quoted an Engineer.ai spokesman as saying that "in a recent app process developed by the company, about 82% of the software was" automatically generated in the first hour, "which is the charm of AI. "

However, Engineer.ai 's internal engineers revealed in an interview with the Wall Street Journal that AI did not automatically assemble code for applications as the company claimed.

They point out that most of the company's work depends on "human engineers" in India and elsewhere. And even given the widespread camouflage mentality of tech startups, the use of artificial intelligence has been exaggerated.

When asked by the media about the company's cases of using artificial intelligence, the company said that the price and project schedule that customers had to pay were calculated completely automatically.

"some of these processes use natural language processing, which is an artificial intelligence technology designed to recognize and understand text or speech.

In addition, you use a decision tree (based on the selected graph or model) to assign tasks to developers. "

However, some current and former employees told the Wall Street Journal that in fact some pricing and schedule calculations are generated by traditional software, and most of the work is generally done manually by employees.

These people even say that the company does not have much natural language processing technology and that the decision trees used within the company should not be regarded as artificial intelligence.

This statement was endorsed by Luka Crnkovic-Friis, founder of Peltarion, a Swedish deep learning software company. He believes that it is often "a little far-fetched" to call decision trees artificial intelligence.

"if you tell customers that you are using artificial intelligence, they may not expect to use some of the technologies of the 1950s. Decision tree is actually a very old and simple technology. "

Interestingly, just this month, Engineer.ai quietly made some more detailed explanations and explanations about his technology and products on the website.

They revised the introduction to "on average, about 60% of the reusable software is produced by machines, and the rest is artificially generated for developing applications."

We are not fully automated application development. Instead, we rely on the partnership between artificial and artificial intelligence, in which reusable software (in traditional software development, repetitive code accounts for about 60% of the product) is machine-produced, and the remaining 40% is machine-produced; what makes most projects unique is "manual production". We believe that human creativity and talent will always be part of the journey of innovation and construction.

In fact, from a practical point of view, it is not good for us to spend a lot of money on fully automated development, and we can realize the interests of our customers by focusing on using automation to solve repetitive and inefficient work (although 80-20 rules are classic, but we are 60-40 rules!).

In addition to the technical application of the product is questioned, it is also considered to follow the "fake it till they make it" principle in the time of launch of the product.

It was only in the past two months that the company began to build the technology needed to build automated applications, according to a person familiar with the company's operations.

He added that the company is at least a year away from applying artificial intelligence technology to its core services.

Of course, as a threshold technology, more and more startups have found it much harder to build artificial intelligence than expected. But in addition to the technology itself, it may also take a long time to collect data to train the machine learning algorithms that support it.

To train the new algorithm model, application developers like Engineer.ai need to collect thousands of requests from customers and combine them with code built by engineers.

But Wall Street quoted several people familiar with engineer AI as saying that the company had not yet collected enough data to support the use of artificial intelligence technology.

But a company spokesman retorted that they had collected more than 600 million records to build a better AI model.

In addition, Robert Holdheim, the former chief business officer of Engineer.ai, who was fired in February and filed a lawsuit against the company (none of which was previously made public), also revealed what founder Duggal had told him:

He says every tech startup exaggerates in order to get capital. I actually agree, which is not surprising. Only in this way can we have the money to develop this technology.

But Duggal has told investors that engineers have done 80% of the development work, but in fact, we haven't started developing the product yet. "

one

Domain name has more than one AI, and financing may be 50% more.

We can't tell whether "fake it till they make it" is right or wrong, and artificial intelligence technology can help companies save money or find target users more accurately in many cases.

But the reality that the investment sector has to face is that in the field of technology, it is an increasingly serious challenge to assess whether a company has made effective use of artificial intelligence technology.

Because artificial intelligence technology itself is complex, vaguely defined and loosely defined, it is difficult for non-professionals to tell when and how it is deployed and used effectively.

In the face of investors, many startups say they are using artificial intelligence as the main way to attract customers, but this argument is often difficult to pass strict scrutiny.

So in general, it is easier to get the love of capital by coming up with a "AI-driven" solution.

The average transaction size of artificial intelligence startups almost tripled from 2013 to 2018, according to CB insights, an authoritative data analytics firm.

PitchBook, another data analytics firm, said venture capital investment in artificial intelligence startups almost doubled in 2018 from the previous year to $31 billion.

In particular, the number of companies whose domain names contain "ai" has more than tripled in a year. At present, this paid domain name extension is very popular among technology start-ups around the world.

In addition, just last month, Japanese technology giant Softbank Corp. Group announced another artificial intelligence technology-based investment fund, Vision Fund II, with an estimated total capital of US $108 billion.

As a two-year-old fund pool, the total amount of Vision Fund I has reached about $100 billion, of which about $70 billion has been injected into a number of AI technology companies.

Engineer.ai, headquartered in Los Angeles, just last year included Deepcore inc. Raised $29.5 million from investment institutions, while Deepcore inc. It is a wholly owned subsidiary of Softbank Corp..

The continuous capital investment of Softbank Corp. and other institutions may continue to improve the market valuations of AI companies on the one hand, and make the skepticism of many technical experts and some investors more unified on the other.

"the biggest problem with artificial intelligence technology so far is that it promises too much, but it can't be achieved," Darrell West, director of the Center for technological Innovation, a think-tank at the Brookings Institution, lamented last week.

"since when, this has become a marketing tool. "

As we just mentioned in the first part, there are clear barriers to the application of this technology.

On the one hand, although it can easily start in a test or preliminary form, it is much more difficult to deploy on a large scale.

On the other hand, obtaining and tagging the necessary training data to build capable artificial intelligence models can be extremely expensive and time-consuming, which is the fundamental reason why a Silicon Valley artificial data tagging company we reported on Monday was able to grow into a unicorn within three years.

However, in view of the poor discrimination of some investors and the limited application of technology, I do not know when entrepreneurs become more addicted to the understanding that "only by using artificial intelligence as a shield can we raise more capital."

A survey of more than 2000 AI technology companies by MMC Ventures, a UK investment fund, found that startups that claim to have some kind of artificial intelligence technology can attract 15-50 per cent more capital than other software companies.

However, they also say that 40% or more of these companies actually do not use any form of artificial intelligence technology at all.

Philipp Gerbert, an artificial intelligence expert at Boston Consulting, argues that startups cannot be blamed.

The strong global interest in artificial intelligence financing and the "technological arms race" among countries have prompted startups and listed companies to begin to advertise themselves as artificial intelligence technology + service organizations.

"even though they may only have a simple chat robot. "

two

Talent is scarce, but AI is growing.

"AI talents" is one of the few hot topics in the technology circle in recent years. However, this topic also leads us to a contradiction that scratches the scalp and doesn't understand:

On the one hand, there is such a shortage of AI talents that major training institutions are starting to sell "AI crash courses". On the other hand, why are companies claiming to have AI technology springing up?

Another problem that is being questioned by Engineer.ai lies in talent.

The Wall Street Journal judged that the company may lack a group of senior staff with deep expertise in machine learning or artificial intelligence.

Because when they were first asked to introduce a senior employee with a background in artificial intelligence technology, they provided only one name.

In a subsequent statement, Engineer.ai also admitted that artificial intelligence experts are really hard to find. But they also say some of the employees recently hired are studying machine learning and artificial intelligence.

However, the company only detailed the experience of three team members in data science and other disciplines in the statement, without specifying their names.

It's easy to think of One Concern, which was questioned by the New York Times last week, which also has talent problems such as a "lack of on-the-job AI technology developers with research results."

Clearly, this is not a corporate dilemma.

According to industry insiders, at present, many companies that claim to have AI technology capabilities will use cheap manpower as a temporary stopgap measure to launch real machine learning algorithms after constantly recruiting people and collecting data.

"A company I know says it is using artificial intelligence software to read and collect receipts, when in fact they are using humans to do the job. It's no secret in the industry. "

Since 2015, the demand for relevant talent has expanded from AI technology to a broader industry, which has led to a surge in demand for employees with artificial intelligence, data science and related field skills.

According to data released in June by CompTIA, a US technology industry group, the unemployment rate in the IT industry fell to 1.3% in May, the lowest level in 20 years.

This intensifies the competition for scarce talent.

Therefore, in the process that this kind of technical talents are more likely to gather with large enterprises and star start-ups, perhaps the consideration of new companies and industrial technology companies is not as difficult to distinguish as foreign media say.

Https://www.toutiao.com/a6725249070530560525/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.