10 Predictions about the Future of Healthcare AI - The Medical Futurist

TRANSCRIPT with timestamps

0:00

after sifting through hundreds of

0:02

studies reports announcements and books

0:05

about artificial intelligence and

0:07

generative AI role in the near future of

0:09

medicine and Healthcare I thought I

0:10

would boil all those down for you into a

0:12

list of 10 predictions I stand by I hope

0:15

that these will serve as a Guiding Light

0:17

for you so you would be in a better

0:18

position to understand where we are

0:20

heading when it comes to using this

0:22

amazing breakthrough technology in

0:24

medicine and Healthcare let's jump right

0:26

into it

0:37

[Music]

Multimodal MLMs

0:40

the future of large language models is

0:43

multimodal jgb like TOS are amazing at

0:47

providing us with the answers to the

0:49

questions we ask through prompts or

0:50

briefs but these large language models

0:52

are usually unimodel meaning they can

0:55

only deal with text some of the newer

0:56

versions like gp4 can now deal with

0:59

images but the next iterations you'll be

1:01

able to deal with video sound and even

1:03

full documents you upload a PDF document

1:06

and you can ask questions about the

1:07

content of it so the large language

1:09

model's future is multimodel meaning

1:12

these mlms will deal with all these

1:14

modalities text image video sound

1:17

whatever kind of formats you want to

1:19

upload you will be able to get the

1:20

answers you're looking for do you know

1:22

why it's important because these mlms

1:25

will be the ultimate interface between

1:27

Physicians using a myriad of different

1:29

different types of AI based Technologies

Annotators

1:32

data annotators will be celebrated there

1:35

are many Physicians right now who are

1:37

sitting in dark rooms manually

1:39

annotating medical images medical

1:41

reports electronic medical records

1:43

without them there is no AI Revolution

1:46

because AI models need a lot of data

1:49

especially annotated data sets to

1:51

improve the more higher quality data

1:54

they have the better predictions these

1:57

models can create so not only we should

1:59

should celebrate the annotators but I

2:01

think they will be cherished and

2:03

rewarded accordingly AI will not replace

Physicians

2:06

Physicians for many years I'm sure

2:08

you've heard the reports and investors

2:10

in interviews saying that this or that

2:12

medical specialty will be vanished and

2:15

that AI will replace medical

2:16

professionals that's simply not and

2:18

cannot be the case not only patients

2:20

need empathy from people they trust and

2:23

that's not Technologies but people human

2:25

beings Physicians but there are so many

2:27

issues and tasks and challenges in a

2:30

medical professional's job that simply

2:32

cannot be automated plus the whole point

2:34

of using AI in medicine is to contribute

2:37

to the process of medical decision

2:39

making so AI will not replace medical

2:41

professionals however I still think that

2:43

those medical professionals that use AI

2:46

will replace those that do not AI will

repetitive and database tasks

2:49

take over repetitive and database tasks

2:52

these are the tasks that represent the

2:54

lwh hanging fruit for the whole

2:56

healthcare industry if a task is

2:58

repetitive and or databased especially

3:01

when combined then that task should not

3:04

be performed by a medical professional

3:06

who has been training for years if not

3:09

decades their human judgment their

3:11

creativity their intuition would not be

3:13

needed to perform that task but AI is

3:17

amazing at doing the same so all these

3:19

repetitive and database tasks especially

3:22

thinking about Administration here could

3:24

and should be automated that's the ho

3:27

Grail of the whole AI revolution in

3:29

medicine AI will find unusual

unusual associations

3:31

associations there have been studies

3:33

showing that artificial

3:35

intelligence-based models and

3:36

Technologies can indeed find biomarkers

3:40

or results in a study that no clinician

3:43

researcher or physician has been

3:45

thinking of these are amazing things

3:47

because they can find associations

3:49

between two end points that we thought

3:52

before to be unrelated such as analyzing

3:55

the phone recordings of patients and

3:57

then coming up with a vocal biomarker

4:00

uh predicting their risk for Alzheimer's

4:02

disease we have never done it before but

4:04

the AI started finding these unusual

4:07

associations especially in the case of

4:09

unsupervised learning models you will

a common language

4:12

need a Common Language with AI to

4:14

understand its progress and I'm not

4:16

talking about learning how to code

4:18

coding is the language of developers

4:21

that make AI possible but you need a

4:23

Common Language such as a board game or

4:25

chess or go something through which you

4:28

would understand a constant prog

4:29

progress AI has been going through for

4:31

me it turned out to be Chess By

4:33

dedicating thousands of hours to chess

4:35

over the last few years I think I've

4:37

been able to better understand what an

4:40

amazing progress AI has experienced in

4:43

those last few years the more I

4:45

understand about the intricacies the the

4:48

tiny nuances of the game I think the

4:50

better I see how far away the best chess

4:53

algorithm is from the bird's absolutely

4:56

best player even Magnus Carson so

4:58

developing a language with AI puts you

5:01

into a better position to understand

5:03

where it's heading prompt engineering is

prompt engineering

5:05

the number one skill in the generative

5:08

AI era chances are that you have come

5:10

across generative AI already by using a

5:13

large language model and as large

5:14

language models like chbt will become

5:17

multimodal dealing with image sound

5:19

video and any kind of data you will have

5:21

to interact with these AI based

5:23

Technologies through the interface of a

5:25

large language model thus the number one

5:27

emerging skill for you now is prompt

5:30

engineering the ability to design the

5:32

absolute best text based briefs or

5:35

prompts to get the best results the best

5:37

outcome you were looking for while using

5:40

these large language models we need

bias

5:42

proper guidelines from medical

5:43

associations about fighting heal equity

5:46

and bias AI cannot be biased by

5:50

themselves they are they can be biased

5:52

because of the database being biased

5:54

that they were trained on therefore this

5:56

must be the conscious preparation and

5:58

decision of the developers to make sure

6:00

that the database they use to train the

6:03

AI on is not biased a good example for

6:06

that is around skin checking

6:07

applications turned out many Studies

6:09

have shown that people with darker skin

6:12

had worse results from the AI based

6:14

algorithms simply because the anatomical

6:17

databases didn't contain enough photos

6:19

and images of skin lesions of people

6:22

with darker skin the database was biased

6:24

so of course the AI started making

6:26

biased decisions but again we can't

6:28

fight against that new regulations will

regulatory category

6:30

be needed generative AI will have to get

6:32

its own regulatory category there is one

6:35

more challenge here that Regulators

6:37

around the world must face now it's

6:38

about adaptive algorithms algorithms

6:41

that will change with every decision

6:43

they make so imagine that there is a

6:44

medical technology that Regulators

6:46

regulate and adopt allow to to be rolled

6:49

out to the market but that technology

6:51

will be a million times different from

6:54

the next day because that's how fast the

6:56

algorithm can change so dealing with

6:58

adaptive algorithms and creating a new

7:00

regulatory category for generative AI

7:03

especially medically trained large

7:05

language models will be the absolute

7:07

priority for Regulators for the coming

7:09

years medicine and Healthcare will

deepfakes

7:11

struggle with deep fakes imagine

7:13

reaching out to you as a healthcare

7:15

service virtually and having a remote

7:17

consultation with a physician but that

7:18

physician is not standing or sitting

7:21

behind a computer there's actually no

7:23

physician behind a computer The Entity

7:25

you're having a video discussion with is

7:27

a deep fake made by AI and the

7:30

discussions you are having with that

7:31

entity is being fueled through a large

7:34

language model like chpt and to find out

7:37

whether it's a deep fake or not through

7:38

a chat based service a chatbot or

7:40

through a video consultation it's going

7:42

to be a bigger and bigger challenge for

7:44

patients while Healthcare Providers will

7:46

try to fill in the gaps left by Dr

7:49

shortages worldwide I hope you found

7:52

these predictions useful and as this is

7:53

an extremely exciting yet dynamically

7:56

changing word I will keep on having a

7:58

finger on the PS of AI and generative AI

8:01

so if any new thing comes up you can

8:03

make sure that I will cover those in the

8:05

upcoming videos

8:07

cheers if you like this video please

8:09

subscribe below to get all the videos

8:11

about the future of medicine Healthcare

8:13

and Advanced Technologies also please

8:15

check out medical futures. think.com to

8:18

access our courses on digital health and

8:21

ai's role in the future of healthcare

8:23

see you there

English (auto-generated)

Artificial Intelligence is Rapidly Evolving

Google looks to show AI's wide opportunities

Google on Tuesday announced a slew of new AI-related products, including new ways it will use its Gemini model to improve products such as Google Search and Google Photos.

Why it matters: Google is using its I/O developer conference to highlight AI advances across the company as it looks to demonstrate it is keeping pace with OpenAI, Microsoft and others while defending its stronghold in search.

Here's what Google has announced so far:

• Google DeepMind head Demis Hassabis unveiled Project Astra, Google's vision for a much more responsive AI-powered assistant that can converse naturally, without lag, while taking in a great deal of information. In one impressive example, Astra was able to help a user find her glasses because they'd shown up in recent video clips on her phone.
• It's bringing some improvements to the paid version of the Gemini app, including a more conversational tone and the ability to interrupt the assistant (features that OpenAI announced for ChatGPT on Monday.)
• Gemini Advanced users will also get the ability to create custom chatbots, known as Gems, somewhat akin to OpenAI's custom GPTs.
• Hassabis also announced Gemini 1.5 Flash, a lighter-weight model designed to be faster and cheaper than the original Gemini 1.5.

In Search, Google is launching its AI overviews — which summarize search results — across the U.S. this week, a move that has publishers worried about an inevitable decline in traffic from the search giant.

• In Photos, Gemini will allow users to ask questions and get answers from their photo collections starting this summer. In a separate demo, Google showed the ability to use a video as part of a search query, in this case to get help using a record player.
• In Gmail, Google showed how AI can help corporate and small business users not only organize their inbox, but also pull out data, such as putting receipts into a spreadsheet. That task can then be automated going forward.
• Google also unveiled a new Veo engine for creating AI-generated videos, which will be available to select creators in the coming weeks; and Imagen 3, an update to its text-to-image engine that can better understand prompts and more capably render text.
• On the Android front, Google said it is making Gemini the mobile software's new built-in assistant and making AI-powered search more prominent, among other enhancements coming this year. It's also expanding a "circle-to-search" feature to help with math and other tasks. And it is testing a feature that would use on-device AI, monitoring phone calls in real time, to alert to potential scams.
• For the data center, Google announced Trillium, its sixth generation Tensor processor chip, available to cloud customers in late 2024. Google also said it will be among the first to offer Nvidia's Blackwell chips, starting in early 2025.

Zoom in: Many of Google's announcements focus on increasing the size of the "context window" — the amount of information the model can use in conjunction with a user's prompts.

• Google already allows developers to use 1 million tokens — roughly hundreds of pages of text or an hour of video — and said it will expand that to 2 million tokens, starting with select developers
• It's also making the 1-million token context available for subscribers of its paid Gemini Advanced service.

Another big focus for Google is transforming AI chatbots into agents that can take action, albeit with human supervision.

• CEO Sundar Pichai shared an example of how someone moving to a new city could use an agent to get information on services in their new neighborhood, change their address across multiple web sites and more.
• "We are thinking hard about how to do it in a way that is private and secure and works for everybody," Pichai said.

Google also unveiled a new watermarking technique for text to identify if it was AI-generated without altering the quality or meaning of the writing.

• Hiding watermarks in text has been a longstanding problem as compared to images, audio, and video where it is easier to add undetectable signals to show that content has been created by AI.
• Google says in its announcement that the new features, which are being added to its existing SynthID tool, will work best with prompts for longer and more varied text, like writing an essay, a theater script or "variations on an email."

The big picture: Google's launches come amid a furious pace of AI announcements across the major tech companies.

• OpenAI on Monday announced its faster, less costly GPT-4o model and a more conversational chatbot.
• Microsoft holds its Build conference next week in Seattle, while Apple is expected to unveil its generative AI strategy at its WWDC event on June 10.
• It has become essential to review and update your AI applications, although most announce updates well in advance.