On October 19th the FDA updated its AI-enabled device list. Being the curious types we are (don't talk about the writing team -Ed.) we downloaded the data and looked at it for interesting insights. 

When we looked at the types of devices being approved radiology and imaging was by far the leader. We charted the types that had 10 or more devices and only one wasn't explicitly imaging-related.

The focus on imaging means that most devices are associated with the radiology specialty; that makes sense. However, to see the extent of this concentration is still surprising.

Next, we wanted to see which companies had the most devices in the list. Again, the focus on imaging wasn't surprising.

The FDA list is actually called the AI/ML devices list. So, when we are able to chart back to 1995 that is why. We ignored the years with zero approvals in the early days and showed the explosion of devices starting in 2018.

2023 isn't complete so we had ChatGPT estimate Q4. Which brings us to our final insight.

For the first chart above we were going to re-learn how to use the Index command in Excel to pull in the code names from the FDA list of over 6,000. Instead, we used ChatGPT Plus with its Data Analysis mode. All the charts in this story were generated by ChatGPT Plus.

MarketingProfs has compiled a list of the five top AI tools in six categories. The categories end up overlapping a lot because "writing" could also be "content" but there are so many AI tools now it's impossible to know them all so having a short list is valuable.

It's important to highlight that OpenAI's DALL-E is now linked to the ChatGPT Plus accounts. This combination has shown itself to be much more contextually aware than MidJourney. Here is an example of each with the prompt: "an oncologist in a well organized office telling a patient about their test results, the patient is sad :: photorealistic style"

The relationship between the actors in the DALL-E image is more nuanced and MidJourney has difficulty knowing which person is the doctor.

Also, DALL-E goes out of its way to show different ethnicities when that is not specified in the prompt. 

Finally, we'll leave you with a tip. This video highlights how to use "Custom Instructions" (only available with ChatGPT Plus) to get even better DALL-E images.

The FDA released a draft guidance last week that expands on its directions regarding off-label communications. In this case it is the supplying of scientific information with HCPs and this document intends to show firms how to stay on the right side of the FDA when doing that.

Some of the information in the guidance about Scientific Information on Unapproved Uses (SIUU) highlights the do's and don'ts:

In the requirements for SIUU is also a link to the most current FDA-required labeling. The FDA wording is "mechanism for obtaining it" which we interpret as a link for digital communications. You can also see a review of the document in RAPS. 

As Google's deadline for deprecating third-party cookies approaches (see our other story in this issue) marketers are looking for solutions. For both advertisers and publishers their own first-party data is the top hope... just not the other's. 

An article from Comparitech looks at the number and scale of cyberattacks on healthcare provider organizations. The numbers have been climbing and even though the last two years have been lower than the peak there is an obvious trend.

We calculated the average patients in each attack and found that had been growing as well. So attacks are getting more common and larger. With the value of patient data one of the highest on the black market that makes logical sense. 

When Comparitech mapped out the attacks they mostly reflect the US population, so cybersecurity seems to be equally bad across all states.

There are many stories about how brands should take a stand on societal issues, especially for Gen Z. The details under the headline are more nuanced and pharma brands should think hard about where and when to do this. 

In the US, support overall for brands talking about issues has reduced from 2022. 

The support for taking a stand on different topics does vary. Climate change leads the list at 55% and Religion trails at 15%. Healthcare is fourth on the list at 49%.

The demographics do follow the traditional trend with Gen Z quite a it higher than the others. 

There are some combinations of topics and demographics that could make taking a stand worthwhile for brands, but by and large it may not be a useful approach for either patient or HCP audiences.

Artists are not happy that AI tools are taking their work, only slightly modifying it, and passing it off a unique work. A new tool called Nightshade is being set up to "poison" AIs and protect artists. 

If you've been ignoring AI topics, here is one of the images that Getty Images is using to show that Stable Diffusion is stealing many of their images. It doesn't even remove the watermark.

A new tool created by researchers looks to poison AI image generation by feeding them images that can distort what is created. In the image below the source image at the top are shown as what the poisoned AI generates at the bottom.

In the above image you can see that Dog becomes Cat and Car becomes Cow.

Will AI models respond if this technology gets adopted? Yes. Likely it will result in what the authors want, which is ignoring poisoned images, but the vast majority of images on the internet aren't protected so the large AI models will likely not be affected.

If you say a company harms children you are guaranteed to trigger an emotional response. Meta is in the news now for a set of lawsuits from 33 states accusing the company of knowingly harming children for profit. 

This adds Meta to Bytedance (TikTok) and Google (YouTube) which already have lawsuits making similar claims.

Any marketers working in Social Media should look to Common Sense Media, an independent non-profit that researches all types of media. They published a report that highlights both the good and the bad of social networks on young people. 

For history buffs, we reported on the whistleblower in 2021 and that story has details from that incident.

Digital health investment is down globally in Q3 2023 according to CB Insights' latest report. The amount invested in Q3 was the lowest since 2016 according to the company.

The distribution remains relatively stable with Asia taking a slightly larger slice of the smaller pie.

Look on the bright side, at least when Google removes third-party cookies from Chrome the FTC won't have anything to complain about regarding medical information.

Notice that the opt-in testing is already open. Google's Privacy Sandbox initiative is looking to help ad-based businesses survive the cookie-pocalypse and has tools available.

Also, while only 1% of browsers will be updated to deprecate third-party cookies in Q1 of 2024, the wording in the story sounds like it will be a ramp to all browsers by Q3. If so marketers need to understand that in Q2 their audiences could hit 50% cookies-disabled and their measurements will already be invalid.

Regardless of our alliteration in the title, this regular quarterly report from Propel shows that journalists are opening slightly more PR emails than they did in 2022.

Forgive the half quote from John Wanamaker, a new study from Claravine looks at the state of marketing data. The company surveyed 140 marketers about the state of their data and estimated that on average 25% of their creative is shown to the wrong audience (half of half?).

Agencies, apparently, were more pessimistic at 30% on average.

The report also looks at who needs to be involved in data standards and here data science / BI comes first.