AI was born way before the Terminator movies in 1984 and iRobot in 2004. It was first described in the 50s in the following paper.
Lots of math involved, but the text in the beginning makes it pretty clear how AI works to solve business problems.
https://www.rand.org/content/dam/rand/pubs/papers/2008/P550.pdf
Moore's Law has finally brought computer processing power fast enough to bring AI to fruition. Right now, in 2024, AI has the brain power of a rat, but by 2025, it is on the trajectory to have the processing power of a human brain.
What are Companies Doing with AI?
AI can do things like help a company optimize a process of when each truck should deliver recycled aluminum so the temperature is just right to create higher yields of new cans. It reduces the need for communication between drivers, schedulers, and the foundry. You can extrapolate that example to: AI can figure out the optimum way to do any business process.
AI for Data Science and Machine Learning
ChatGPT.com is a website we've all heard a lot about. ChatGPT can suggest to programmers what python coding libraries to use for a job. It can look at a block of python code and find syntax errors. If you've never done coding, finding errors can be very tough. It can convert R to Python code and other codes as well.
ChatGPT can summarize a long article quickly which is something most of us can use.
I'm taking certificate courses on AI now to understand it and use it's power to do marketing better-faster-cheaper, so my employers can use the productivity gains for competitive advantage.
If everyone else is using it, that means I will just be running two steps behind if I'm not.
AI can save a data analyst or a software programmer twenty hours of work a week. It has definitely saved me hours and hours of work already.
AI Helped Me with Marketing, Needs Work on Linkedin
I used Claude.ai to help me create an ebook for my employer to use as a giveaway for a tradeshow recently. I was having trouble pulling good factoids on the high costs of drug-and-alcohol abuse in the workplace; Google was giving me nothing, OSHA gave me almost nothing, but Claude.ai pulled up three pages of relevant facts in about sixty seconds. That really saved me because at startups we wear a lot of hats and need to get a lot done quickly.
Microsoft"s Powerpoint "Design Studio" feature program uses machine learning to propose choices of relevant graphics and page layouts for each page of my ebook related to my topic, and even matched the text colors I used for our brand. I created it as slides then just flipped it to portrait. I've been using that design feature in powerpoint for a couple of years now and love the results.
Linkedin offered me, and everyone else, AI to write cover letters to recruiters, when I was applying for jobs. What AI came came up with was very wordy and repetitive. I would be embarrassed to use that text. It also mentioned my relevant experience in the cover letter, but always pulled my most recent experience for every position, not what was relevant to the position I was applying for. It did give ideas for elements of what should be included, which is helpful. You can tell an engine like Claude.ai or Chat GPT to do it again shorter or with a more specific focus if something comes out wordy, but with Linkedin's feature I just edited it quickly.
How to Get The Best Results Writing text with AI
The key to getting good results with AI is to be really specific when asking Claude.ai or any other AI engine to do something, just like when you are searching via google. For instance, you could type in: "five-minute speach to a group of Salespeople about 'elevate'" and it will print out a speach; Or if you are in finance, you can type in "manual for forecast procedures at a software company" or "tax laws for revenue recognition of sales, 2024". Of course, you always have to verify you are getting good info, but AI gives you a great jumping off point and very quickly.
How AI Works in a Nut Shell
So to describe how AI works my course talks about how a human would teach a mechanical dog to walk, as in, tell it to move front right and back left legs forward then front left and back right; With AI, you have to let the computer in the dog try things, it learns the most efficient way by rating results of steps for walking (simplistically) +1 or -1, or -5 if it falls, then it optimizes "walking" to an even greater efficiency.
Gaming AI Technology-Learning by Temporal Differences
My current favorite game app on my iphone, chess.com. It let's me play chess any time against the computer and is powered by AI. I read that I will never be able to beat it at the Master level. I rarely beat it at the advanced level. Taught two friends visiting me to play this weekend. One flew home to California and invited me to play on the app the next day, so fun!
Summarized how AI powers many games using Claude.ai:
"Temporal Differences learning is a model-free method, meaning it does not require explicit knowledge of the transition probabilities or reward function of the MDP. It can learn directly from experience, making it suitable for problems where the environment dynamics are unknown or too complex to model explicitly.
Scientists Use Models from Nature to Improve on AI
Thought this fact was interesting, scientists modeled neural nets based on how ring worms do them, because nature is more efficient in 2020 and still using these models to update in 2023.
What are liquid neural nets?
Liquid neural nets use a new mathematical formulation and wiring pattern to create deep learning models that are compact, energy-efficient, and causal. It can address some of the key challenges of current deep learning models and create new directions for AI research.
- The inspiration for LNNs was to create machine learning models that can run on robots and other resource-constrained edge devices.
- LNNs use a mathematical formulation that is less computationally expensive than traditional ANNs and stabilizes neurons during training.
- LNNs also use a wiring architecture that is different from traditional neural networks and allows for lateral and recurrent connections within the same layer.
- Rus and her colleagues were able to train an LNN with just 19 neurons to perform a task such as keeping a car in its lane the same task—a traditional ANN would require ~100,000 neurons and ~500,000 parameters.
- With so few neurons, LNNs are much more interpretable and we can extract a decision tree that corresponds to the firing patterns and essentially the decision-making flow in the system.
- Experiments show that LNNs learn causality and focus on the task instead of learning spurious patterns in the environment.
- One characteristic to take note of is that LNNs only work with time series and sequential data—you can’t use them for static datasets such as ImageNet."
AI is Speeding Up Delivery of Media and Networks
I'm posting this article with Google AMP (Accelereated Mobile Pages) so it will load instantly and keep readers reading longer according to Hubspot.
Wondering if Google bought that tech from Salesforce.com who bought it from TwinPrime where I worked in 2016? We used machine learning, protocols, and infrastructure to speed up delivery of images for customers with apps in verticals that use a lot of images like social media, gaming, and etailers.
It worked by trying to send the images in different ways in peak traffic times in densely populated areas, then AI would figure out the best way to get them through quickly, in example, one at a time might go faster than trying to send every image in this article at once.
Looked it up, AMP is definitely powered by AI as I suspected.
Found this online on Readwrite.com: "Furthermore, the AI-powered AMP can facilitate online content and e-commerce businesses to enhance digital marketing strategies, increase the rankings, and better mobile-friendly experience."
Talked to a Team Leader at Cisco yesterday where they are using AI to improve ethernet networking.
Summary
There are thousands, if not millions, of ways that AI is being used to create new and faster technology, translate code, locate and analyze data, write text for specific use cases, optimize business processes, speed up networks, create slides, games and more. You still have to carefully scrutinize AI output for edits, but productivity gains can be immense when used properly.
Just hoping we don't run into a Terminator / iRobot situation, along with a lot of other people, ha.
Sources:
- Learning to Predict by the Methods of Temporal Differences, by Richard S. Sutton
- Tech Talks 2020 and 2023, Scientists Use Models from Nature to Improve on AI
- Udemy Certificatuion Course: Artificial Intelligence A to Z, 2024
- ReadWrite.com Google Accelerated Mobile Pages
