Strategic resources for AI/automation?
August 26, 2023 6:02 AM Subscribe
I work in healthcare IT, and have always been on the side of requesting data rather than the one writing the queries. I’m looking for some leadership level resources that I can use to help guide conversations away from the hype and into the realm of possibility.
Most of what I found in my first pass presumed I was looking to learn to code (this knowledge would be helpful but we have much more technically proficient people available) or was breathless, likely AI-generated articles light on specifics (“this API works with many popular programming languages to reduce inefficiency”). Are there books or other resources that would be helpful other than Peter Lee’s excellent AI Revolution in Medicine? Should I do a CS course or certificate just to get some hands on experience and more easily translate what I’m seeing in GitHub into conversations?
Most of what I found in my first pass presumed I was looking to learn to code (this knowledge would be helpful but we have much more technically proficient people available) or was breathless, likely AI-generated articles light on specifics (“this API works with many popular programming languages to reduce inefficiency”). Are there books or other resources that would be helpful other than Peter Lee’s excellent AI Revolution in Medicine? Should I do a CS course or certificate just to get some hands on experience and more easily translate what I’m seeing in GitHub into conversations?
I would want to know what's the source of data you're hoping to nab.
The field of getting data in the right shape for consumption is known as data science, and there are many courses on Coursera, linked to job titles such as "data scientist", "data analyst", even "data engineer". I remember taking data science courses off of Coursera and one of the applications they talk about was specifically how to use the data in the medical field. The field was been in obscurity until relatively recently when computing power and databases have reached the point where it's now a popular field, but as a buzzword it was soon overtaken by AI.
I personally haven't seen a lot of talk about how to link AI with data science, but I can imagine it will help you spit out some data science-y code using R or Python and the processing library such as Numpy and Scikit to massage the data into the shape you need.
posted by kschang at 8:38 AM on August 26, 2023
The field of getting data in the right shape for consumption is known as data science, and there are many courses on Coursera, linked to job titles such as "data scientist", "data analyst", even "data engineer". I remember taking data science courses off of Coursera and one of the applications they talk about was specifically how to use the data in the medical field. The field was been in obscurity until relatively recently when computing power and databases have reached the point where it's now a popular field, but as a buzzword it was soon overtaken by AI.
I personally haven't seen a lot of talk about how to link AI with data science, but I can imagine it will help you spit out some data science-y code using R or Python and the processing library such as Numpy and Scikit to massage the data into the shape you need.
posted by kschang at 8:38 AM on August 26, 2023
If you are just hoping to "scrape" the data off existing web pages and whatnot, and has a modicum of intelligence via AI, there's a tool that "scripts" interactions called Bardeen.ai that builds "automations" (scripts) that will run and do things for you, that used to require an IT guy to actually code the script, now can be done via a few clicks.
Some of the scenarios they outline were "FIND new product/person/news; ADD row to Google Sheets; NOTIFY person via Slack". It sounds like a smarter version of IFTTT or Zapier. Instead of YOU personally doom scrolling, cutting and pasting, have the "AI" do it for you.
posted by kschang at 8:52 AM on August 26, 2023 [1 favorite]
Some of the scenarios they outline were "FIND new product/person/news; ADD row to Google Sheets; NOTIFY person via Slack". It sounds like a smarter version of IFTTT or Zapier. Instead of YOU personally doom scrolling, cutting and pasting, have the "AI" do it for you.
posted by kschang at 8:52 AM on August 26, 2023 [1 favorite]
used to require an IT guy
Women work in IT, too.
posted by soelo at 8:54 AM on August 26, 2023 [7 favorites]
Women work in IT, too.
posted by soelo at 8:54 AM on August 26, 2023 [7 favorites]
The field of getting data in the right shape for consumption is known as data science
God I hope not! My team of data scientists are almost entirely PhDs. Data science is more ML modeling or inference. Yes we munge data, but that’s like saying the field of writing git commit messages is known as data science.
I don’t think a CS course will be useful, that’s going to be mostly theoretical. What you may want is data manipulation and munging? AI can definitely make that smoother: asking a LLM. “I want this table in R from go to wide to long” will spit out the relevant code.
posted by MisantropicPainforest at 9:31 AM on August 26, 2023 [3 favorites]
God I hope not! My team of data scientists are almost entirely PhDs. Data science is more ML modeling or inference. Yes we munge data, but that’s like saying the field of writing git commit messages is known as data science.
I don’t think a CS course will be useful, that’s going to be mostly theoretical. What you may want is data manipulation and munging? AI can definitely make that smoother: asking a LLM. “I want this table in R from go to wide to long” will spit out the relevant code.
posted by MisantropicPainforest at 9:31 AM on August 26, 2023 [3 favorites]
[I'm a bit confused, because these responses don't seem to be addressing the question asked. So maybe I'm missing something and am off-topic myself.]
"AI" is a broad topic that covers many different technologies. You might need to separate those out, determine which are most relevant to you, and focus on one at a time. Large language models (LLMs), for example, are the hot AI topic of the moment everywhere. There are other forms of AI finding application in medicine, though (e.g., computer vision systems for radiology). But as at least part of an answer, I'll talk about LLMs here.
If you want to be able to have informed conversations about LLMs and to help guide others in those conversations, I think you might do well to start by learning about how LLMs work and their strengths and weaknesses. Once you have a foundation in that, then you might dig deeper into medicine- and healthcare-specific applications, as you will be better able to assess claims you find there. This article, for example, might be the kind of thing that helps at first. (It looks like one of the authors has a newsletter/podcast called Understanding AI that may also be useful.)
You can also augment that by playing around with LLMs yourself to gain a better sense of their abilities and limits. You can use both OpenAI's ChatGPT and Anthropic's Claude model for free once you create an account with each. Ask them factual questions, probe the limits of their knowledge in your area, ask reasoning questions... slowly you'll get a sense of what they can and cannot do.
Any introductory CS course will be too low-level to inform conversations about LLMs or other types of AI systems. You'd have to take several classes (five or so?) before that might start to inform conversations you have about AI. You'll definitely want to stay at a higher level, focusing on using and applying AI technologies with maybe a bit about how they are created, how they are trained, and how they work internally.
posted by whatnotever at 2:19 PM on August 26, 2023 [3 favorites]
"AI" is a broad topic that covers many different technologies. You might need to separate those out, determine which are most relevant to you, and focus on one at a time. Large language models (LLMs), for example, are the hot AI topic of the moment everywhere. There are other forms of AI finding application in medicine, though (e.g., computer vision systems for radiology). But as at least part of an answer, I'll talk about LLMs here.
If you want to be able to have informed conversations about LLMs and to help guide others in those conversations, I think you might do well to start by learning about how LLMs work and their strengths and weaknesses. Once you have a foundation in that, then you might dig deeper into medicine- and healthcare-specific applications, as you will be better able to assess claims you find there. This article, for example, might be the kind of thing that helps at first. (It looks like one of the authors has a newsletter/podcast called Understanding AI that may also be useful.)
You can also augment that by playing around with LLMs yourself to gain a better sense of their abilities and limits. You can use both OpenAI's ChatGPT and Anthropic's Claude model for free once you create an account with each. Ask them factual questions, probe the limits of their knowledge in your area, ask reasoning questions... slowly you'll get a sense of what they can and cannot do.
Any introductory CS course will be too low-level to inform conversations about LLMs or other types of AI systems. You'd have to take several classes (five or so?) before that might start to inform conversations you have about AI. You'll definitely want to stay at a higher level, focusing on using and applying AI technologies with maybe a bit about how they are created, how they are trained, and how they work internally.
posted by whatnotever at 2:19 PM on August 26, 2023 [3 favorites]
I work in healthcare IT, ON the side of writing the queries, and I agree that playing around with python or whatever isn't going to be that helpful if your question is more along the lines of "how do I help my organization make good decisions about using or not using AI."
I'm not sure what search terms you're using, but there's a lot (I mean A LOT - it is totally unavoidable right now) written about how AI is being used in medicine specifically and health care more generally.
From the professional journals:
A nice overview of the history and future of AI in medicine
Another nice overview of the current and future of health care AI
NEJM is starting a journal devoted to AI in medicine
Take a course (not on Python!):
A Coursera course from Stanford on AI in Healthcare
The consultant class weighs in:
McKinsey on generative AI in healthcare
PwC on AI in healthcare generally
Some items from the health care IT press:
A look at heath care AI in 2023
An interview with a physician in the health care AI industry
Other:
A social perception view from Pew
I would also see what Harvard Business Review and Gartner have on the topic. If your org has any relationships with the big consulting firms, it may be worthwhile to ask for white papers on the topic as well.
I do feel like your question is a little confusing. You're asking about leadership-level, strategic information for AI, which indicates you're being asked to decide or give input on decisions to spend the org's time and treasure on AI. This kind of decision-making isn't compatible to poking around on GitHub or taking an entry-level CS course. There's not anything wrong with doing either of those, but they also won't help you actually make decisions. It might be worthwhile to partner with someone in your org with more of a technologist background to assess the potential for your situation and use cases.
If you've just got a contingent of folks overly excited about something they found on GitHub, consider letting them run a small pilot with it (if you can do some compliantly given your org's data and regulatory requirements) to get it out of the realm of hype.
posted by jeoc at 6:44 PM on August 26, 2023 [1 favorite]
I'm not sure what search terms you're using, but there's a lot (I mean A LOT - it is totally unavoidable right now) written about how AI is being used in medicine specifically and health care more generally.
From the professional journals:
A nice overview of the history and future of AI in medicine
Another nice overview of the current and future of health care AI
NEJM is starting a journal devoted to AI in medicine
Take a course (not on Python!):
A Coursera course from Stanford on AI in Healthcare
The consultant class weighs in:
McKinsey on generative AI in healthcare
PwC on AI in healthcare generally
Some items from the health care IT press:
A look at heath care AI in 2023
An interview with a physician in the health care AI industry
Other:
A social perception view from Pew
I would also see what Harvard Business Review and Gartner have on the topic. If your org has any relationships with the big consulting firms, it may be worthwhile to ask for white papers on the topic as well.
I do feel like your question is a little confusing. You're asking about leadership-level, strategic information for AI, which indicates you're being asked to decide or give input on decisions to spend the org's time and treasure on AI. This kind of decision-making isn't compatible to poking around on GitHub or taking an entry-level CS course. There's not anything wrong with doing either of those, but they also won't help you actually make decisions. It might be worthwhile to partner with someone in your org with more of a technologist background to assess the potential for your situation and use cases.
If you've just got a contingent of folks overly excited about something they found on GitHub, consider letting them run a small pilot with it (if you can do some compliantly given your org's data and regulatory requirements) to get it out of the realm of hype.
posted by jeoc at 6:44 PM on August 26, 2023 [1 favorite]
As a strategic decision maker, maybe there is a benefit in reading more broadly about how AI is being applied in different industries and sectors? I recently added this MIT Press book featuring case studies of AI to my reading list.
posted by amusebuche at 7:22 PM on August 26, 2023 [2 favorites]
posted by amusebuche at 7:22 PM on August 26, 2023 [2 favorites]
You are not logged in, either login or create an account to post comments
posted by Probabilitics at 6:51 AM on August 26, 2023