Customer service chatbot sector forecast to be worth $7bn this year

Plus: You can now turn photos of your dead relatives into talking, blinking deepfakes

In brief Conversations with chatbots are still pretty clumsy today, but they're improving – especially in customer service – and could be worth big money.

The machine learning software powering these services is allowing companies to tailor general customer service chatbots on their own data. These agents can provide more helpful suggestions instead of generic responses.

For example, the financial unit of General Motors that deals with things like offering car loans, has a chatbot powered by IBM Watson Assistant to help customers pay their loans. The automated software has been programmed to understand certain queries, such as "What is my payoff amount?" or "Did you receive my March payment?" according to the New York Times.

The text-based digital assistant has saved the company an estimated $935,000; it doesn't have to pay for human workers or lengthy phone calls. Gartner reckons the industry for these customer service chatbots will grow 15 per cent to more than $7bn this year.

Business is booming. The technology is rapidly improving; there are a more startups providing this type of software and lots of companies looking to buy.

'No silver bullet for responsible deployment'

OpenAI is going to release new datasets that test the safety of language models like GPT-3.

These systems can generate text that is classified as offensive, NSFW, and false. Companies like OpenAI are trying to figure out how to make language models safer to use so they can be adopted more widely. At the moment, they rely on content filters.

The San Francisco-based business admitted there is currently no silver bullet for responsible deployment, and that it should have been more careful with regard to the types of data GPT-3 ingested during the training process.

Some developers using the GPT-3 API have tried to abuse it, getting it to generate text to promote dodgy medical products or racist fantasy roleplays. OpenAI has developed datasets to test the safety of its models, and is going to release them.

"Specifically, we have developed new evaluation metrics for measuring toxicity in model outputs and have also developed in-house classifiers for detecting content that violates our content policy, such as erotic content, hate speech, violence, harassment, and self-harm," it said.

Bring your deceased loved ones to life with deepfakes

Dead relatives appear to speak and move in AI deepfakes created by a startup in partnership with MyHeritage, a genealogy company focused on helping customers flesh out their family trees.

You can see some examples here. All you need to create your own so-called LiveStory clip is to upload a photo of your loved one, and write a short passage of text describing their life. You can edit the voices to match what they sound like too. The results aren't quite perfect, sometimes the mouth movements don't quite match up to the speech; the facial expressions and blinks are also a bit unnatural.

"Our use of AI to breathe new life into historical photos is unique and is helping millions of people cultivate a renewed emotional connection with their ancestors and deceased loved ones," Gilad Japhet, founder and CEO of MyHeritage, said in a canned statement. "Genealogy is all about telling and preserving our family stories. We keep showing the world how fun and compelling genealogy can be."

The software behind the LiveStory feature was developed by D-ID, a startup that builds a range of AI-based creativity tools, and licensed by MyHeritage.

Waymo's self-driving cars to make grocery deliveries

Waymo has been granted a permit by California Public Utilities Commission to start charging customers for ordering trips made by its self-driving cars in San Francisco.

The permit doesn't cover taxi services, only deliveries, according to TechCrunch. Waymo has partnered with Albertsons, an online grocery store, to deliver goods to people in San Francisco. A human driver must also be present at all times.

Riders can't sit in one of Waymo's flashy AI robo-vehicles with or without any human supervision yet. "We take a step-by-step approach on the path to rolling out our fully autonomous experience to the public," Nick Smith, a Waymo spokesperson, said. 

"We start with an autonomous specialist behind the wheel operating in autonomous mode, and open the rides to a select group of Trusted Testers for free, before we begin charging. Eventually we move to launching in rider-only mode (without anyone else in the car)."

Waymo did not disclose how much it will be charging for deliveries. Meanwhile, Cruise launched a public waitlist for folks to sign up and become its first users. ®