How To Talk To Robots

The 2000s

Fei-Fei Li (1976– ) & ImageNet

Fei-Fei Li worked at the intersection of computer science and neuroscience for years before founding ImageNet in 2006. ImageNet is an enormous database of labelled images used for object recognition. It was set up by Li because she recognised AI needed a database as well as algorithms and formulae. Building ImageNet through meetings with Christiane Fellbaum – one of the leaders of WordNet doing the same for language – it is now considered absolutely invaluable for AI research. ImageNet led to the deep learning revolution (the period where deep learning became the standard in AI) by tweaking training algorithms to make them perform much better. Essentially, it changed the AI game.

What I admire so much about Fei-Fei Li is the way she has always been invested in making AI more accessible and diverse, as well as considering its impact. In 2011, she started the non-profit A14ALL, which runs summer programs in AI for women, people of colour and low-income high-school kids.

Watson IBM Jeopardy

Remember IBM’s Deep Blue chess match? Well in 2010, IBM developed a machine that was trained by AI to play the game Jeopardy, learning the right and wrong answers from a bank of old Jeopardy questions. They named it Watson. The machine had no access to the Internet, so couldn’t search the answers online. Instead, it had to figure out the answers from what it already knew. It was part of a research plan to make the public more interested in AI and its potential. In 2011, Watson played two human contestants and won the game.

Voice Assistants

Another big moment in the use of everyday AI was the introduction of voice assistants. Siri was the first virtual AI assistant that could not only combine speech recognition and information retrieval but could also live conveniently in our pockets. Apple acquired it from its namesake company in 2010, then launched it as part of the iPhone the following year. In 2015, Amazon brought in Alexa, a device that responds to questions in real time, like Siri, but is activated by the sound of its name, rather than the push of a button. It was initially marketed as a way to make life easier: scheduling a calendar, playing music, connecting to other useful apps. But now Alexa can connect to smart home devices to switch lights on and off, and you might even hear her manipulated to rap on records.

DeepMind, Alibaba, Tencent, Baidu & OpenAI

DeepMind is a start-up AI research company founded in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman in London. DeepMind is now run by a seriously awesome woman, Lila Ibrahim, and is a subsidiary of Alphabet, the most cash-rich company in the world (it even owns Google!). Like the chess games of the 1990s, the DeepMind team have been applying and developing AI techniques to the game Go – an ancient Chinese two-player boardgame, established over 3,000 years ago. In their computerised version, named AlphaGo, the number of possible board configurations is, to quote the DeepMind website, ‘More than the number of atoms in the known universe.’ By employing a whole range of AI techniques including reinforcement learning, AlphaGo won its first match against a human in 2015 before going on to win the Go summit in 2017. One of the wildest things is that after that 2017 summit, DeepMind released AlphaZero, which unlike IBM Watson was never taught from a trained dataset. Instead, it learned how to play from trial and error against itself, starting with totally random moves. This creates a way to win that human beings almost never do themselves. The DeepMind team see this as a critical step towards AI-augmented human creativity, and the potential of Artificial General Intelligence. Demis Hassabis is recognised as one of the biggest brains of his generation, so watch this space …

In the same year that AlphaZero was released, the Chinese government started pouring millions of dollars into AI development. Their focus was on helping organisations, companies and the military adapt and deploy AI, as well as integrating AI training into education. The Chinese government has plans to have built a $1 trillion AI industry by 2030. There are three major tech companies in China leading this AI revolution: Alibaba, Tencent and Baidu. These companies are described by journalists and governments alike as rivals to the American giants Amazon and Google. And what they’re building definitely challenges the idea that the US is ‘winning the AI race.’ In fact, a recent book by Kai-Fu Lee called AI Superpowers examines how the USA and China are developing AI, and argues that China has the upper hand. For example, the Chinese company Alibaba is currently the world’s largest e-commerce marketplace and has invested in seven AI research labs that will focus on machine learning, network security and natural language processing. And in 2018, Alibaba released robots that outperformed humans on the Stanford University reading test.

OpenAI

OpenAI is the AI research company founded by, among others, Elon Musk. In February of 2019, OpenAI created language software that was so good at learning human speech patterns and then generating text that they deemed it too risky to release the code to the public. This is because it could be misused, particularly for fake news and spamming people. But it’s important to note that currently machines can’t write perfect prose, nor can they have a true understanding of the world in order to be manipulative. What we need to be conscious of is that the people who are controlling them can’t either.

In August 2019, OpenAI decided to release a more curtailed version of the modeller that you can actually play around with yourself without having to use code. Try entering some text and seeing what the OpenAI bot refers back with by visiting https://talktotransformer.com, entering a prompt that the AI will complete.

I fed the machine Spice Girls lyrics, and below is the result. I’ll leave it to you to decide if we should have AI Spice on the reunion tour.

Prompt: ‘2 Become 1’, by the Spice Girls

Completion: Now come in here and show me how it’s done Now I’m watching a little history By the end of the night, I’ll know what’s going on And that, all I need are some deep thoughts I need some love.

When reading the history of AI, it’s important that we consider that the ideas of ‘winning’, ‘leading’ and ‘progressing’ in AI are not synonymous with ‘better’. We’ll explore more about this in Chapter Four, but good questions to keep in mind as we’re looking at more modern advancements are: what does it mean to be winning this race? What would you prioritise as a measure of success? What is the prize? And who gets to claim it and has been everybody been acknowledged? And even: who gets to decide what it means to be a leader of AI?

Конец ознакомительного фрагмента.

Текст предоставлен ООО «ЛитРес».

Прочитайте эту книгу целиком, купив полную легальную версию на ЛитРес.

Безопасно оплатить книгу можно банковской картой Visa, MasterCard, Maestro, со счета мобильного телефона, с платежного терминала, в салоне МТС или Связной, через PayPal, WebMoney, Яндекс.Деньги, QIWI Кошелек, бонусными картами или другим удобным Вам способом.