Connect with us

Technology

The technology behind OpenAI’s fiction-writing, fake-news-spewing AI, explained

https://www.technologyreview.com/s/612975/ai-natural-language-processing-explained/

Published

on

Last Thursday (Feb. 14), the nonprofit research firm OpenAI released a new language model capable of generating convincing passages of prose. So convincing, in fact, that the researchers have refrained from open-sourcing the code, in hopes of stalling its potential weaponization as a means of mass-producing fake news.

While the impressive results are a remarkable leap beyond what existing language models have achieved, the technique involved isn’t exactly new. Instead, the breakthrough was driven primarily by feeding the algorithm ever more training data—a trick that has also been responsible for most of the other recent advancements in teaching AI to read and write. “It’s kind of surprising people in terms of what you can do with […] more data and bigger models,” says Percy Liang, a computer science professor at Stanford. 

The passages of text that the model produces are good enough to masquerade as something human-written. But this ability should not be confused with a genuine understanding of language—the ultimate goal of the subfield of AI known as natural-language processing (NLP). (There’s an analogue in computer vision: an algorithm can synthesize highly realistic images without any true visual comprehension.) In fact, getting machines to that level of understanding is a task that has largely eluded NLP researchers. That goal could take years, even decades, to achieve, surmises Liang, and is likely to involve techniques that don’t yet exist.

Four different philosophies of language currently drive the development of NLP techniques. Let’s begin with the one used by OpenAI.

#1. Distributional semantics

Linguistic philosophy. Words derive meaning from how they are used. For example, the words “cat” and “dog” are related in meaning because they are used more or less the same way. You can feed and pet a cat, and you feed and pet a dog. You can’t, however, feed and pet an orange.

How it translates to NLP. Algorithms based on distributional semantics have been largely responsible for the recent breakthroughs in NLP. They use machine learning to process text, finding patterns by essentially counting how often and how closely words are used in relation to one another. The resultant models can then use those patterns to construct complete sentences or paragraphs, and power things like autocomplete or other predictive text systems. In recent years, some researchers have also begun experimenting with looking at the distributions of random character sequences  rather than words, so models can more flexibly handle acronyms, punctuation, slang, and other things that don’t appear in the dictionary, as well as languages that don’t have clear delineations between words.

Pros. These algorithms are flexible and scalable, because they can be applied within any context and learn from unlabeled data.

Cons. The models they produce don’t actually understand the sentences they construct. At the end of the day, they’re writing prose using word associations.

#2. Frame semantics

Linguistic philosophy. Language is used to describe actions and events, so sentences can be subdivided into subjects, verbs, and modifiers—who, what, where, and when.

How it translates to NLP. Algorithms based on frame semantics use a set of rules or lots of labeled training data to learn to deconstruct sentences. This makes them particularly good at parsing simple commands—and thus useful for chatbots or voice assistants. If you asked Alexa to “find a restaurant with four stars for tomorrow,” for example, such an algorithm would figure out how to execute the sentence by breaking it down into the action (“find”), the what (“restaurant with four stars”), and the when (“tomorrow”).

Pros. Unlike distributional-semantic algorithms, which don’t understand the text they learn from, frame-semantic algorithms can distinguish the different pieces of information in a sentence. These can be used to answer questions like “When is this event taking place?”

Cons. These algorithms can only handle very simple sentences and therefore fail to capture nuance. Because they require a lot of context-specific training, they’re also not flexible.

#3. Model-theoretical semantics

Linguistic philosophy. Language is used to communicate human knowledge.

How it translates to NLP. Model-theoretical semantics is based on an old idea in AI that all of human knowledge can be encoded, or modeled, in a series of logical rules. So if you know that birds can fly, and eagles are birds, then you can deduce that eagles can fly. This approach is no longer in vogue because researchers soon realized there were too many exceptions to each rule (for example, penguins are birds but can’t fly). But algorithms based on model-theoretical semantics are still useful for extracting information from models of knowledge, like databases. Like frame-semantics algorithms, they parse sentences by deconstructing them into parts. But whereas frame semantics defines those parts as the who, what, where, and when, model-theoretical semantics defines them as the logical rules encoding knowledge. For example, consider the question “What is the largest city in Europe by population?” A model-theoretical algorithm would break it down into a series of self-contained queries: “What are all the cities in the world?” “Which ones are in Europe?” “What are the cities’ populations?” “Which population is the largest?” It would then be able to traverse the model of knowledge to get you your final answer.

Pros. These algorithms give machines the ability to answer complex and nuanced questions.

Cons. They require a model of knowledge, which is time consuming to build, and are not flexible across different contexts.

#4. Grounded semantics

Linguistic philosophy. Language derives meaning from lived experience. In other words, humans created language to achieve their goals, so it must be understood within the context of our goal-oriented world.

How it translates to NLP. This is the newest approach and the one that Liang thinks holds the most promise. It tries to mimic how humans pick up language over the course of their life: the machine starts with a blank state and learns to associate words with the correct meanings through conversation and interaction. In a simple example, if you wanted to teach a computer how to move objects around in a virtual world, you would give it a command like “Move the red block to the left” and then show it what you meant. Over time, the machine would learn to understand and execute the commands without help.

Pros. In theory, these algorithms should be very flexible and get the closest to a genuine understanding of language.

Cons. Teaching is very time intensive—and not all words and phrases are as easy to illustrate as “Move the red block.”

In the short term, Liang thinks, the field of NLP will see much more progress from exploiting existing techniques, particularly those based on distributional semantics. But in the longer term, he believes, they all have limits. “There’s probably a qualitative gap between the way that humans understand language and perceive the world and our current models,” he says. Closing that gap would probably require a new way of thinking, he adds, as well as much more time.

This originally appeared in our AI newsletter The Algorithm. To have it directly delivered to your inbox, sign up here for free.

Click to comment

You must be logged in to post a comment Login

Leave a Reply

Technology

These ten enterprise M&A deals totaled over $40B in 2019

It would be hard to top the 2018 enterprise M&A total of a whopping $87 billion, and predictably this year didn’t come close. In fact, the top 10 enterprise M&A deals in 2019 were less than half last year’s, totaling $40.6 billion. This year’s biggest purchase was Salesforce buying Tableau for $15.7 billion, which would…

Published

on

These ten enterprise M&A deals totaled over $40B in 2019

It would be hard to top the 2018 enterprise M&A total of a whopping $87 billion, and predictably this year didn’t come close. In fact, the top 10 enterprise M&A deals in 2019 were less than half last year’s, totaling $40.6 billion.

This year’s biggest purchase was Salesforce buying Tableau for $15.7 billion, which would have been good for third place last year behind IBM’s mega deal plucking Red Hat for $34 billion and Broadcom grabbing CA Technologies for $18.8 billion.

Contributing to this year’s quieter activity was the fact that several typically acquisitive companies — Adobe, Oracle and IBM — stayed mostly on the sidelines after big investments last year. It’s not unusual for companies to take a go-slow approach after a big expenditure year. Adobe and Oracle bought just two companies each with neither revealing the prices. IBM didn’t buy any.

Microsoft didn’t show up on this year’s list either, but still managed to pick up eight new companies. It was just that none was large enough to make the list (or even for them to publicly reveal the prices). When a publicly traded company doesn’t reveal the price, it usually means that it didn’t reach the threshold of being material to the company’s results.

As always, just because you buy it doesn’t mean it’s always going to integrate smoothly or well, and we won’t know about the success or failure of these transactions for some years to come. For now, we can only look at the deals themselves.

Continue Reading

Technology

Jumia, DHL, and Alibaba will face off in African ecommerce 2.0

The business of selling consumer goods and services online is a relatively young endeavor across Africa, but ecommerce is set to boom. Over the last eight years, the sector has seen its first phase of big VC fundings, startup duels and attrition. To date, scaling e-commerce in Africa has straddled the line of challenge and…

Published

on

Jumia, DHL, and Alibaba will face off in African ecommerce 2.0

The business of selling consumer goods and services online is a relatively young endeavor across Africa, but ecommerce is set to boom.

Over the last eight years, the sector has seen its first phase of big VC fundings, startup duels and attrition.

To date, scaling e-commerce in Africa has straddled the line of challenge and opportunity, perhaps more than any other market in the world. Across major African economies, many of the requisites for online retail — internet access, digital payment adoption, and 3PL delivery options — have been severely lacking.

Still, startups jumped into this market for the chance to digitize a share of Africa’s fast growing consumer spending, expected to top $2 billion by 2025.

African e-commerce 2.0 will include some old and new players, play out across more countries, place more priority on internet services, and see the entry of China.

But before highlighting several things to look out for in the future of digital-retail on the continent, a look back is beneficial.

Jumia vs. Konga

The early years for development of African online shopping largely played out in Nigeria (and to some extent South Africa). Anyone who visited Nigeria from 2012 to 2016 likely saw evidence of one of the continent’s early e-commerce showdowns. Nigeria had its own Coke vs. Pepsi-like duel — a race between ventures Konga and Jumia to out-advertise and out-discount each other in a quest to scale online shopping in Africa’s largest economy and most populous nation.

Traveling in Lagos traffic, large billboards for each startup faced off across the skyline, as their delivery motorcycles buzzed between stopped cars.

Covering each company early on, it appeared a battle of VC attrition. The challenge: who could continue to raise enough capital to absorb the losses of simultaneously capturing and creating an e-commerce market in notoriously difficult conditions.

In addition to the aforementioned challenges, Nigeria also had (and continues to have) shoddy electricity.

Both Konga — founded by Nigerian Sim Shagaya — and Jumia — originally founded by two Nigerians and two Frenchman — were forced to burn capital building fulfillment operations most e-commerce startups source to third parties.

That included their own delivery and payment services (KongaPay and JumiaPay). In addition to sales of goods from mobile-phones to diapers, both startups also began experimenting with verticals for internet based services, such as food-delivery and classifieds.

While Jumia and Konga were competing in Nigeria, there was another VC driven race for e-commerce playing out in South Africa — the continent’s second largest and most advanced economy.

E-tailers Takealot and Kalahari had been jockeying for market share since 2011 after raising capital in the hundreds of millions of dollars from investors Naspers and U.S. fund Tiger Global Management.

So how did things turn out in West and Southern Africa? In 2014, the lead investor of a flailing Kalahari — Naspers — facilitated a merger with Takealot (that was more of an acquisition). They nixed the Kalahari brand in 2016 and bought out Takelot’s largest investor, Tiger Global, in 2018. Takealot is now South Africa’s leading e-commerce site by market share, but only operates in one country.

In Nigeria, by 2016 Jumia had outpaced its rival Konga in Alexa ratings (6 vs 14), while out-raising Konga (with backing of Goldman Sachs) to become Africa’s first VC backed, startup unicorn. By early 2018, Konga was purchased in a distressed acquisition and faded away as a competitor to Jumia.

Jumia went on to expand online goods and services verticals into 14 Africa countries (though it recently exited a few) and in April 2019 raised over $200 million in an NYSE IPO — the first on a major exchange for a VC-backed startup operating in Africa.

Jumia’s had bumpy road since going public — losing significant share-value after a short-sell attack earlier in 2019 — but the continent’s leading e-commerce company still has heap of capital and generates $100 million in revenues (even with losses).

Continue Reading

Technology

Airbnb’s New Year’s Eve guest volume shows its falling growth rate

Hello and welcome back to our regular morning look at private companies, public markets and the gray space in between. It’s finally 2020, the year that should bring us a direct listing from home-sharing giant Airbnb, a technology company valued at tens of billions of dollars. The company’s flotation will be a key event in…

Published

on

Airbnb’s New Year’s Eve guest volume shows its falling growth rate

Hello and welcome back to our regular morning look at private companies, public markets and the gray space in between.

It’s finally 2020, the year that should bring us a direct listing from home-sharing giant Airbnb, a technology company valued at tens of billions of dollars. The company’s flotation will be a key event in this coming year’s technology exit market. Expect the NYSE and Nasdaq to compete for the listing, bankers to queue to take part, and endless media coverage.

Given that that’s ahead, we’re going to take periodic looks at Airbnb as we tick closer to its eventual public market debut. And that means that this morning we’re looking back through time to see how fast the company has grown by using a quirky data point.

Airbnb releases a regular tally of its expected “guest stays” for New Year’s Eve each year, including 2019. We can therefore look back in time, tracking how quickly (or not) Airbnb’s New Year Eve guest tally has risen. This exercise will provide a loose, but fun proxy for the company’s growth as a whole.

The numbers

Before we look into the figures themselves, keep in mind that we are looking at a guest figure which is at best a proxy for revenue. We don’t know the revenue mix of the guest stays, for example, meaning that Airbnb could have seen a 10% drop in per-guest revenue this New Year’s Eve — even with more guest stays — and we’d have no idea.

So, the cliche about grains of salt and taking, please.

But as more guests tends to mean more rentals which points towards more revenue, the New Year’s Eve figures are useful as we work to understand how quickly Airbnb is growing now compared to how fast it grew in the past. The faster the company is expanding today, the more it’s worth. And given recent news that the company has ditched profitability in favor of boosting its sales and marketing spend (leading to sharp, regular deficits in its quarterly results), how fast Airbnb can grow through higher spend is a key question for the highly-backed, San Francisco-based private company.

Here’s the tally of guest stays in Airbnb’s during New Years Eve (data via CNBC, Jon Erlichman, Airbnb), and their resulting year-over-year growth rates:

  • 2009: 1,400
  • 2010: 6,000 (+329%)
  • 2011: 3,1000 (+417%)
  • 2012: 108,000 (248%)
  • 2013: 250,000 (+131%)
  • 2014: 540,000 (+116%)
  • 2015: 1,100,000 (+104%)
  • 2016: 2,000,000 (+82%)
  • 2017: 3,000,000 (+50%)
  • 2018: 3,700,000 (+23%)
  • 2019: 4,500,000 (+22%)

In chart form, that looks like this:

Let’s talk about a few things that stand out. First is that the company’s growth rate managed to stay over 100% for as long as it did. In case you’re a SaaS fan, what Airbnb pulled off in its early years (again, using this fun proxy for revenue growth) was far better than a triple-triple-double-double-double.

Next, the company’s growth rate in percentage terms has slowed dramatically, including in 2019. At the same time the firm managed to re-accelerate its gross guest growth in 2019. In numerical terms, Airbnb added 1,000,000 New Year’s Eve guest stays in 2017, 700,000 in 2018, and 800,000 in 2019. So 2019’s gross adds was not a record, but it was a better result than its year-ago tally.

Continue Reading

Recent Posts