Thinks and Links | June 14, 2024
BD&A - Thinks and Links
Big Data & Analytics - Thinks and Links | News and insights at the intersection of cybersecurity, data, and AI
đź“« Subscribe to Thinks & Links direct to your inbox
Happy Friday!
Getting Personal
The biggest AI story of the week is the news from Apple’s Worldwide Developer Conference about the integration of ChatGPT into iPhones in the next version of the OS. This is incredible in light of the facts that apple has had Siri on iPhones since 2011, has reportedly spent over $100 billion in R&D over the last five years, and OpenAI is significantly funded and enabled by Microsoft. It is remarkable to see a deal like this go through – who could have guessed that these two would be working so closely together 18 years ago?
While the OpenAI + Apple discussion got much of the attention, it’s worth noting that many of the new features that are coming to iOS 18 do not involve ChatGPT and many will not even leave the phone. This is the real future of AI that is going to be worth paying attention to. Much of it will remain on the device, using smaller models that can handle simpler tasks. When needed, the devices will reach out to Apple’s cloud for other AI models to perform tasks in secure and confidential computing “secure enclaves.” Requests will go to OpenAI when they can’t be done by either prior method. I would guess that over time if Apple can handle more of the requests on their own, they will. So this may be a temporary arrangement.
But since 2024 is the year to provide transformative AI features to users, and this is likely what will happen with AI when you update to iOS 18:
On-Device AI:
Text analysis and response suggestions
Categorization and prioritization of mail and notification
Basic image processing and document editing
Create custom emojis
On Apple’s Cloud
Create movies from photos based on your description
Summarize notifications, news articles, and voice memos
Some of the advanced Siri features
Sent to OpenAI
Re-writing, proofreading, and summarizing text
Image understanding
Siri questions and complex searches
Siri will also get much smarter in ways that will likely have to use OpenAI models to be successful. It will be able to bring context into responses by remembering previous requests and having awareness of what’s on your screen. It will be able to read from and execute commands into various apps on the phone. It will be able to answer more complex question structures. If no privacy and security constraints are on it, there should be very little on the iPhone that Siri can’t do. You respond to emails in perfect grammar while driving a car. You can have an assistant that books a trip to a foreign country, invites your friends, and translates your preferences into local language instructions. You can write a newsletter, including images and links, and have it send on your behalf.
But privacy and security are very important. Time will tell how secure Apple’s cloud compute is. With 17% of the global smartphone market, and a generally more affluent user base than other platforms, these datacenters are a target rich environment. If the AI data can be intercepted between phones and datacenter or if the cloud itself is breached, there’s going to be a lot of information about people’s personal lives, their finances, their businesses, and more. Likewise, while OpenAI is promising to not link iOS requests to IP address and not retain the data it processes, it adds additional steps and widens the attack surface. If an attacker manages to instruct Siri with all of this advanced capability, it could be devastating for the iOS users who are impacted.
The push is on to bring advanced AI features to users, and IT and Security teams must take action. Unlike the Microsoft Copilot laptops announced a few weeks ago, an iOS update can arrive almost instantaneously to a huge portion of the population. Approved or not, updating iPhones will invite more AI and OpenAI in to your business. Businesses that use device Mobile Device Management tools such as Microsoft Intune, Scalefusion, and Apple Business Essentials should compare the planned actions iPhones can take against company policy. You can encourage employees not to upgrade for a time, but you will very soon need to have a plan for users with iOS 18 and advanced AI capabilities in your environment.
What comes next? Ask Siri…
Total Recall
https://www.wired.com/story/total-recall-windows-recall-ai/
Microsoft’s new Recall AI tool, announced a few weeks ago and set to launch in June, is raising significant privacy and security concerns. Recall takes screenshots of all user activity on a PC every five seconds and saves that unencrypted data locally. It did not take long for security researchers to weigh in on how that might be a bad idea. Introducing TotalRecall, developed by ethical hacker Alex Hagenah. Within a week of the announcement of Recall, here is a demonstration of how easily a malicious actor could extract all that sensitive data. The tool can pull this data from the unencrypted Recall database in seconds
Security researchers are sounding the alarm, comparing Recall to spyware or a “built-in trojan.” They warn the feature could be exploited by hackers, cyber criminals, or even abusive partners to gain a complete record of a victim’s private digital life. Experts are calling on Microsoft to rework or delay the feature to address these serious vulnerabilities before release. As of last week, Microsoft appears to be listening and setting Recall to be opt-in by default rather than opt-out. But add this to the list of vulnerabilities to monitor closely in future hardware roll outs.
Adobe Wants Your Data
https://9to5mac.com/2024/06/06/change-to-adobe-terms-amp-conditions/
Adobe recently updated its terms and conditions for apps like Photoshop, requiring users to agree to the new terms to continue using the software. The changes have ignited fierce backlash from professionals across various industries who are concerned about the implications for their intellectual property rights. Under the revised terms, Adobe states that it may access user content through automated and manual methods for content review purposes. The company also claims a non-exclusive, worldwide, royalty-free, sublicensable license to use, reproduce, display, distribute, modify, create derivative works, perform, and translate the content. This broad language has alarmed many users who fear Adobe could exploit their work without consent or compensation. It is also symptomatic of need for technology companies to get more data to make better models.
AI Antitrust
The U.S. Justice Department and Federal Trade Commission are taking a closer look at Microsoft, OpenAI, and Nvidia over recent investments and activities that could be anti-competitive. It is likely Apple will join the fray with the latest WWDC AI announcements. These companies have a massive advantage given the kind of business they do. Whether this is illegal or what the impact would be beyond fines is yet to be seen. Investigations like this are likely to take longer than the roll out of new features. It’s unclear what the impact will be in the long term, but for now it is full steam ahead at all of the impacted companies.
What’s in AI name?
https://www.nature.com/articles/d41586-024-00797-z
Researchers have used Machine Learning on Elephant calls to identify personalized sounds that correspond with members of their group. By using recordings of wild elephants, machine learning models were able to identify which sounds corresponded with which animal. The findings confirm the hypothesis that elephants use names, but more research is needed to understand if other information is conveyed in the sounds.
Trust me, I’m old - Experience is Needed to Navigate AI Risks
A recent study of AI trainings found that when junior staff train managers on the use of Generative AI technologies, the risk mitigation tactics are often contrary to expert advice. Less experienced consultants focused on behavior change rather than system design. Top-down governance and risk approaches recognize that complex transformations are hard and require approaches that assume people won’t just follow the instructions. It is critical for seasoned leaders to become familiar with these emerging technologies and risks to implement holistic approaches.
The Prompt Report: An AI-led Review of the Best GenAI Prompting Techniques
https://trigaten.github.io/Prompt_Survey_Site/
A comprehensive study of nearly 5,000 research papers that involved AI review and researcher scrutiny to arrive at 58 text-based techniques and their relative effectiveness for getting Generative AI to do what we want it to do. This work prepared a taxonomy of prompting techniques and the associated research around best practices. Useful resource if you’re tuning your prompting techniques.
Have a Great Weekend!
You can chat with the newsletter archive at **https://chat.openai.com/g/g-IjiJNup7g-thinks-and-links-digest