Monday, May 27, 2024

Amazon CTO: AI will soon become ‘culturally aware’

Werner Vogels, chief technology officer of tech giant Amazon, said in an exclusive interview with Newsbytes.PH that as artificial intelligence (AI) continues to evolve, AI will go beyond what we know today and will soon be “culturally aware.”

Here’s an excerpt from our interview with Dr. Vogels.

Culturally aware AI and how different it is from the AI that we know today

Vogels: Without insulting the current state of the art of AI, let me first do a little bit of history. I make a line between what I would call old-fashioned AI and modern AI. So, if I look at the old-fashioned AI, those are all things that work really well.

And that has worked really well for a long time. Image recognition, video processing, object detection, forecasting, speech-to-text, and natural language understanding, all fields of AI, that work really well. One of the original founders of artificial intelligence is a professor called McCarty. And he had a really famous quote that says, as soon as it works, we don’t call it AI anymore.

Recommendations, similarities, review management, summarizing reviews, all of these things are AI that we’ve been using for the past 20 years and they work really well.

Generative AI and search

Vogels: What changed in the past, let’s say, nine months or a year and a half is that we’ve gone through new software phases and we got to a point where we developed some software technology that is then able to take an enormous amount of text and then make relationships between the words. If you have a table and in the context of the table is a glass and a spoon, then you know it’s a physical table. If columns and rows are in the context, it’s probably a database table. But it’s still very text-oriented.

So this new form of AI, which is called generative AI, because it can generate new text, is most of the text or imagery or video or sounds is really mostly focused on helping people discover things.

It can be a good replacement or an augmentation for search, for example. If you think about the consumer services at this moment in that space, it can be a good replacement for search. Because if I say, oh, what are the best things I could be doing? If my search on any search engine would be, what are the best things to do in Manila this weekend? I probably have to visit five different websites to understand which ones are the commercial ones, which are tricking me into doing something or things like that.

It’s not like this takes over our brains. We’re not there yet with doing complex tasks and things like that. So, most of the innovations in this space started in the United States, of course, with companies like Amazon and Google and OpenAI and Anthropic. And there are dozens and dozens of companies that have done this.

But they’ve trained it mostly on Western data. There’s a very large data set called Common Crawl and that contains every possible type of information and not only facts. A lot of opinions. If you include all data from Twitter and from Reddit and whatever, there are some facts in there, some useful facts, but there’s also a lot of things that may not be true or whatever. (And so, you still need to use your brain. The problem is that all of that is English.

Actually, the Common Crawl is largely English. Even if it’s content from China or from here, it will be still in English. And even though it looks like these systems have different language interfaces, they translate the language into English and then go into the system and do it in English.

So, it’s very, very Western-oriented, those big models. If we want to ensure that the whole world has access to this technology and want to make sure that they have the relevant information, not everything needs to become like America or Western Europe. We treasure our local cultures.

The rise of culturally aware AI

Vogels: And I think, especially here in Southeast Asia, you have so many brilliant different cultures, even within the Philippines. Different islands will have different cultures and different histories. If you ask an American large language model about that, they probably come up with some sort of atlas or history lesson or whatever.

They won’t know about the culture that is embedded often in your languages. Now, if you think about it, I don’t know who the most famous Filipino author or poet is, but if you would ask an American LLM, you get a very different answer about a review of one of his books than you would get if you would have a local knowledge base. It’s not just about the language.

It is the culture of the language that is embedded into it. And as such, I think it’s really important to ensure, to make sure that everybody in the world has access to this tech technology that we build more and more culturally aware LLMs.

Now, of course, we probably have, I don’t know how many cultures there are in the world. I mean, there’s 210 countries. I think there are 7,500 languages in the world. Probably, let’s do that times 10 and you get into the different cultures.

So, there’s 70,000, 100,000 cultures in the world. You can’t have 100,000 LLMs and pick one each. However, what you can have is have them work together. So a Filipino-based or a Japanese-based or a Taiwanese or an Indonesian one can collaborate with others in the Asian region, other LLMs, or with the American or with the Western European LLMs to, together in collaboration, come with the right answer, given the context. Context in all of this is still important.

The role of language

Vogels: And so, there’s many use cases where, you know, whether it’s creativity, idea generation, summarization. You know, if you would have a, how shall I say, imagine you have a very large Filipino text. And you would use the current LLMs to summarize that. It would translate it into English, do a summary, and then translate it back into Tagalog. And so, I’m pretty sure that’s a very different summary.

And so English still is the core between them. And the English language is very unique. Just as Spanish is unique, just as Tagalog is unique, just as all languages are unique.

That’s why we have these different languages. And as such, we should also have technology that serves them in the right way. And the reason why I’m hammering on that is that I think it is extremely important that, you know, this is very fast-moving tech technology.

This is a technology that wasn’t in the hands of consumers a year and a half ago. And we need to make sure that all the kind of immense creative impact on mostly white-collar jobs and things like that. We need to make sure that it’s not just the Western world that is benefiting from that.


- Advertisement -spot_img




- Advertisement -spot_img