AI news - Fri 11th Aug 2023
While it’s been a little quiet this week in terms of new models, or any world changing announcements, it has been a busy week with lots of new real world applications of AI. Unfortunately, these real world use cases come with a lot of worry and negativity. Therefore, we have a lot of shorter (one sentence) stories this week, and as the video conferencing company Zoom have appeared in AI news stories twice this week we are going to focus on them.
Zoom - Hackers could steal our password through Zoom.
Zoom - Are Zoom using our meetings to train AI? They say no, at least not anymore.
Company to watch - Align.ai - trying to combine best in class AI with what humans actually want it to do (Fortune article on the very interesting founder)
An interesting read - Nature have written a useful article on how different nations are regulating AI (Read)
An interesting read - Is the UK ill equipped to protect workers against AI? (Read)
An interesting watch - The founder of Khan academy and Bills Gates talking about AI and education (Watch)
Good to know - Big Chinese tech companies have rushed to buy NVIDA chips needed for AI. This is because there is a concern the US will stop exporting them (Read more)
Good to know - The UK Government have announced two experts to lead the AI Safety Summit (included Matt Clifford from Entrepreneur First) and a £13 million fund to modernise healthcare (Read more)
Good to know - Meta (formerly Facebook) have gone all in on AI, actually folding a department (focused on protein folding, bad joke) to reallocate the funds into AI development(Read more)
Good to know - Disney has created a specific AI task force, who will be looking into how the company can save money (Read more)
AI in real life - AI can now price house using visual clues (Read more)
AI in real life - A shopping chatbot in New Zealand gave shoppers disgusting recipes, including one for chlorine gas (Read more)
AI in real life - An author has found 5 books under her name on Amazon which she claims were written by ChatGPT (Read more)
AI in real life - AI is helping pilot take paths which avoid those white lines in the sky (contails), therefore reducing the impact on our planet (Read more)
Are Zoom Listening?
You may or may not be wondering why the hell a video conferencing company would be using AI. If you actually think about some of Zoom's features, particularly auto captioning, they have to be using new technologies and a lot of user data (i.e. customer conversations) for these to be effective.
There has been an increased interest in the ethics around using customer data to train company algorithms. As well as how much we as consumers should be sharing with these technologies. This opens up a useful conversation around these technologies, for example it seems to be general consensus that we should not be telling ChatGPT about our personal and health issues (ChatGPT itself has a disclaimer about this), but we do not have the same worries about typing these into Google Search.
Companies like Netflix, Spotify and Amazon clearly use our data to recommend products and to try to keep us using the platform which doesn't seem to have such worries attached. Google docs also predict our next words, and this has to be trained on our input (potentially private documents). So why are we so worried about Zoom (if we even are)?
Zoom Have Updated Terms and Conditions
On Monday, Stackdiary called out an update to Zoom's terms and conditions which allowed user data in AI training with no opt out.
Zoom responded by publishing a blog to "confirm that we will not use audio, video, or chat customer content to train our artificial intelligence models without your consent".
However, Zoom are using two new generative AI features - Zoom IQ Meeting Summary and Zoom IQ Team Chat Compose - and whether or not your data is used in these features is not quick so easy as you saying yes or no. These are being offered on a free trial, and account owners opt in. Other participants will receive a notification if the host of a specific meeting opts in.
"Your content is used solely to improve the performance and accuracy of these AI services. And even if you chose to share your data, it will not be used for training of any third-party models."
Interesting choice of wording - nice to know they're not selling our data to the likes of OpenAI to further develop their model. Zoom are just using our data to train their own AI models.
Zoom's blog on their AI assistant - read more
Zoom's blog on their data usage policy - read more
BBC article about Zoom using our data - read more
Research Says Passwords Can Be Heard Through Zoom
If that wasn't enough news on Zoom this week, researchers have also been studying whether or not specific keyboard strikes can be picked up on video call. Their conclusion is yes.
While this sounds unimportant, our founder Laura has experience working in fraud for a bank and she found this worrying. Fraudsters are extremely smart and convincing. Getting a potential victim on a Zoom call (as service like Wix offer as a real form of customer support), and being able to record every key strike during the call makes it very easy to determine a password.
Now let's add AI into the mix and this sounds like a potential recipe for chaos.
Full academic paper on Arxiv - read more
Guardian article about the findings - read more