The thread about iNat potentially using AI to potentially pick out the useful info hidden away in comments got me thinking.
How do iNatters use AI now in scientific ways in their every day lives? When do you find it to be reliable?
I loathe how every search engine now provides an AI summary because I find 90% of the time there are huge inaccuracies if not hallucinations.And now I have to click on every source document to ensure it is relevant. So I end up reading more.
I work in health. I have found the Consensus app to be really helpful for patient management issues. Maybe because you need to frame your question in a scientific manner.
I do write numerous reports but I would never use AI to help write these. Some things are onerous but the difficulties often bring out the issues because they force you to think. I also wonder about the confidentiality of the information if I used AI. And the implicit bias evident in AI responses doesnāt need to be added to my own biases.
I probably have more research in my everyday life than most iNatters, but ChatGPT is very helpful for troubleshooting software and sometimes writing code. I do a fair amount of analysis in R and QGIS for my research, and AI is immensely helpful for this.
Sometimes when I am looking for information and coming up empty, in desperation Iāll try to ask an LLM to give me the information + its sources so I can verify. Unfortunately, this usually ends up being at best just about as effective as what I was doing with a traditional search engine, and rarely gives me anything new to work with if I was already struggling to find something.
I have used it to find papers that I already knew existed but had a hard time re-finding myself, by describing what I remembered about the paper and asking it to find and link it for me. Sometimes what I remember is just imprecise and fuzzy enough that a traditional search engine has a hard time bridging my queries to the actual paper, but the LLMās language processing manages to figure out what Iām looking for.
If by āAIā you mean LLMs and the like, I never use them for anything. For me, the only potential benefits would come if I never bothered to vet the model outputs and the potential risks from not vetting model outputs makes it untenable.
If you mean typical ML approaches, I use them every day as tools to either find approximate answers or filter through huge amounts of data in a coarse manner (iNatās CV model is a great example of this; if I have no idea what organism an observation contains I can at least see what the model predicts for successive images and kind of triangulate to the Family or Genus level).
The only thing I currently personally use LLMs for is for helping out if I need to do minor coding.
Personally, Iād rather just find a code library online somehow, but unfortunately that has become almost impossible to do of late because the tech-knowledge side of the internet is absolutely inundated with garbage LLM-generated junk webpages.
So I just use the LLM to get a starting point on the code I need - I figure, I donāt speak computer fluently but the LLM does so I can use it to bridge that gap. But I also am going in and double checking the code myself and usually making my own edits.
EDIT: I do utilize the CV to be honest. But itās a different sort of program and much less intrusive than LLMs
As Google and other search engines have declined in quality Iāve been finding LLMs more and more useful for doing basic searches for information, for better or for worse. Perplexity AI has been my go-to since itās good at referencing sources and generally accurately summarizes them (better than the LLM summaries included in search engines). It picks out the relevant info much faster than I could in scanning through search engine results. This includes identification comparisons, it seems to do those fairly well and is handy in often including relevant images as well.
For anything in academic papers though usually I need to do my own searching.
At my work only Microsoft Copilot is approved for corporate use, and other LLM websites are blocked on corporate devices for that reason.
Thatās a valid concern. I believe most responses to LLMs are saved and used to ātrainā the software, which is to say, not confidential in the slightest.
In DuckDuckGo you can disable the AI summary in settings, which is my primary reason for choosing that as my default search engine. I donāt want to see it, and I donāt want energy wasted generating it.
I never intentionally use LLMās except one time chatting with one just kind of to see what it was/how human it could seem. That was before I knew about the horrible environmental cost, so Iām glad I was unimpressed and never tried it again.
The only time I āuseā them now is accidentally, when I start reading an article supposedly written by a real person, and get a paragraph or two in before realizing that itās just AI slop. Total waste of time and very annoying!
I do sparingly use Google lens to look up photos, and find it accurate about half the time. Which is still better than nothing, if an object is completely unfamiliar to me. It can also sometimes find more photos of a person out there on the internet, which is useful in genealogy. Although itās not 100% accurate, as evidenced by this result on a photo of my 4xgreat-grandmother that was so funny I saved a screenshot:
the environmental cost is likely to improve over time. possibly at a different rate than mooreās law but for the same basic reasonā¦. human ingenuity.
AIs are just tools. the more you use them, the better you understand what each one is capable of, and the more useful they can be when you really need them.
when working with chatbots, iāve found that the key to getting accurate information is just to ask the chatbot what the source of a given interesting nugget of information is. if the chatbot gives you a reasonably trustworthy source, and you can find that nugget in that source, then great. if not, try asking for alternative sources, ask your questions in different ways, or investigate other interesting nuggets.
these days, i would say that looking for information this way via AI will turn up better results more quickly than using an old-school search engine. thereās just too much clutter in conventional search results from the entire web.
because i know how to code little bit, i donāt use AIs as much for coding because i like to structure my code in particular ways, and in some ways, i enjoy the problem solving challenge involved. but if i just need to do some one-time rapid prototyping or visualization, thereās no harm in spending a few minutes to see if an AI can do it before spending much more time to do it myself. while AIs are exceptionally good at coding these days, i would always caution folks to not rely entirely on AI-generated code for something that is really important without at least doing a code review to make sure itās really doing what you asked for.
regarding confidentiality, at some level you have to reveal what youāre looking for to someone to get an answer. so i think the question is really who you trust to reveal that information to. in a future world where the average person could load an LLM onto their own machine that they control and run it from there, that situation might actually provide you the most privacy, but we will never get to such a world without helping to advance the technologies required to do that.
bias is similar. any source will have baggage and blind spots. but these days, for many areas of knowledge, machines have potential for greater breadth of knowledge than any single human, and theyāre catching up to the capability of collectives of people, too.
i think itād be a really bad idea to rely on AI to come up with a ui flow or design that considers human factors thoughtfully enough. syntax, modern standards, a question about a browser version handling a particular thing etcā¦ AI is really helpful.
I think I must have become a dinosaur. The whole concept of AI, as I understand it, is terrible. I have disabled AI summaries in my browser. I would rather use my own human intelligence than get input from some stupid machine which has no idea what it is talking about. Articles written by AI really get my blood boiling.
I recently moved to a country I do not know the language of. It makes using traditional search enginges less helpful, if you want to find local businesses and such, which are not advertizing in english for example. ChatGPT helped me for example find websites for finding a new flat .. and then automated translators helped getting along there..
Machine translation is a form of AI, right? In trying to compile the current state of knowledge on the genus Achlys, I found that there was very little in English on Achlys japonica. Machine translation allowed me to access materials in Japanese. The translation in such cases doesnāt have to be perfect as long as I get the information accurately.
TBH āAIā as most people are usually using the term is, in my opinion, isnāt super accurate. Itās a blatant misnomer at worst since not of these systems are ACTUALLY intelligent; at best itās a useful catch-all for āautomated computer stuffā since most of the populace doesnāt really care enough to get into the specificities.
I looked it up though and it does seem like Google Translate has moved to a LLM system in recent years, which would bring it in line with what most people are referring to as āAI.ā https://en.wikipedia.org/wiki/Google_Translate - and this is actually a great use for it IMHO. Translation apps are huge and allow people to communicate in ways that they never would have been able to before
Sorry, no idea.
In my professional everyday work as a chemist and material engineer in a large company, i have found yet no useful example for AI use in generating more or better knowledge or in explaining often complicated scientific content in a better way to the audience.
I see few young colleagues letting AI generate code snippets to overcome software gaps, generated due to constant cost-down efforts of the employer. Not a sustainable way of business process if you ask me. I guess it mostly occurs because money is systematically drained away from the real economy through the stock exchange to AI server farms.
The same young colleagues also use AI for adhoc summaries of larger pdf documents and tables from the internet.
But as nobody has the time to verify the source data nor what AI has made with it, it is not considered serious data and is not taken as a baseline for business decisions.
Some managers used AI for writing employee ratings and feedback. Nobody, including those managers, considers it as an amelioration to the work relationship, only as a shortcut to cumbersome social interaction.
The best: A good colleague, PhD physicist, and active member of a local IT club lately did the 180Ā° turn from an open minded AI observer to a complete refusal of AI. He dropped all engagement with Amazon, Meta and Google and changed fully back to linux system and AI free server based social networks. He considers the internet already today as dead as dried cod.
If i remember the outcome of my google search last sunday to get a cost comparison of house emergency call service for my mother, i have to agree to him : The first five non-advertised results showed superficially different websites, all following the same scheme with clearly same automated database behind them. All completely non-serious, phishing after 2nd layer for a telephone number to trick elderly people into expensive contracts.
Fazit: Complete AI enshitification.
Even my wife now woke-up after actively using Alexa for more than a year.
I hope I can send Alexa soon to where it belongs.
Take care. Take your choice. Enjoy the show if you can.
