What about AI?

Consensus was great when it provided abstract information for relevant journal articles, but they have updated their access, and now you get search results that pull straight from the full text of the papers.  In some ways, this will be great.  We all know abstracts rarely provide the information you want.  With the new full-text access, you won’t have to search through a 120 page article to find the relevant parts.  Consensus does that for you.  Of course, it is only a program and is working from your prompts, so some tweaking will definitely be required.  It is also a good idea to pull the full articles to make sure the AI did not miss anything, but this is another tool that will save you enough time to make it worthwhile.

Data Analysis and Prediction

On the surface, these may not seem like student tasks, but analyzing statistics and being able to make informed choices based on your findings is exactly what students should be able to do.

• Tasks: Extracting insights from data, identifying patterns, and forecasting future trends.
  • Predictive Analytics:  Forecasting future events (e.g., customer churn, sales forecasting, fraud detection).
  • Data Cleaning and Preparation:  Automating the process of making data ready for analysis.
  • Anomaly Detection:  Identifying unusual patterns or outliers in data (e.g., fraudulent transactions).
  • Business Intelligence (BI) Augmentation:  Enhancing traditional BI with AI-driven insights.
  • Risk Modeling:  Assessing and quantifying financial or operational risks.

• Programs/Platforms:
  • Google Cloud AI Platform (Vertex AI, BigQuery ML): Comprehensive suite for building, deploying, and managing ML models, including predictive analytics.  These are really big picture programs for creating entirely new AI programs.       
  • AWS SageMaker:  Fully managed service for building, training, and deploying machine learning models.
  • Azure Machine Learning: Microsoft's cloud-based platform for ML.
  • TensorFlow & PyTorch:  Underlying frameworks for building custom predictive models.
  • R:  Statistical programming language widely used for data analysis and machine learning.
  • Databricks:  Data platform that integrates AI/ML capabilities for large-scale data analysis.

Natural Language Processing (NLP)

These are your big, broad tools that do just about anything.  This is where good prompting skills come into play.

• Tasks: Understanding, interpreting, and generating human language.  This includes:
  • Text Summarization:  Condensing long texts into shorter, coherent summaries.
  • Sentiment Analysis:  Determining the emotional tone of text (positive, negative, neutral).
  • Chatbots/Conversational AI:  Enabling human-like conversations, answering questions, and providing support.
  • Translation: Converting text from one language to another.
  • Named Entity Recognition (NER):  Identifying and classifying entities like names, organizations, and locations.
  • Topic Modeling:  Discovering abstract topics in a collection of documents.
  • Text Generation:  Creating new text, like articles, emails, or creative content.

• Programs/Platforms:
  • Google Cloud Natural Language API (GEMINI):  Pre-trained models for sentiment analysis, entity extraction, content classification, and more.  This is Gemini.  This is free all the time and is a pretty good tool.  I use it every day. 
  • OpenAI's GPT series (ChatGPT):  Excellent for general text generation, summarization, and conversational AI.       There is a free level of this.
  • NLTK (Natural Language Toolkit):  A popular Python library for research and development in NLP, offering tools for tokenization, stemming, tagging, parsing, and more.  Python can do anything, but it is a language you will need to learn first.

After going through the big AI answer, I realized that there is no way I will be discussing seven types.  Right off the top, I am going to drop all the programs that write AI programs.  If you have a specialty need for your business, these are great, but this blog is meant for students, staff, and faculty of a community college.  When you are in graduate school, you might need those tools. When you are ready, do an AI search, prompting it to name the tool that will best suit your task.

Another thing I dropped are multiple programs from the same company, specifically Amazon, Microsoft, and Google.  They show up in the list multiple times with different names.  As an example, Amazon has AI programs called Rekognition, Comprehend, and SageMaker and none of them are free.  Essentially, they are all the same – programs that write programs.  All Microsoft programs come with associated costs.  Google has at least nine programs.  They boil down to Gemini and a variety of prompting techniques. 

Python is an open-source programming language that is closer to human language than any other programming language.  It is really flexible and, for our purposes here, can be used to write just about anything.  It is a first-level skill you would have to acquire, like the ability to write HTML.  From there, the possibilities are endless. 

The robotics, healthcare, and financial services modeling tools are simply unrelated to our needs.  I suspected that when I got started, but decided to leave them in, thinking there might be something useful.  There probably is, but for this blog, they are just clutter.  My target audience may never need tools like that.  Once a profession is chosen, the employer will have a solution that fits their needs. 

Computer vision tools are pretty cool, but are pretty deep for our needs.  This is image manipulation, OCR reading, and facial recognition.  You may use these tools a lot in life but probably not much in school.

So, after a false start, over the next four weeks we will take a deep dive into:

1. Natural Language Processing (NLP)
2. Generative AI (Art, Music, Code, etc.)

3. Data Analysis and Prediction

   This is another reason why you should always check your AI response.  It may not fit your needs.

    

My biggest concern with AI is which program to use for what thing.  I have sort of covered that here by naming Consensus as a great AI for student research because it only searches open-source, peer-reviewed journals.  Alternatively, something like Gemini is great for proofreading and general information.  I used it yesterday to reduce a 3000-word story down to 2500 words to meet submission requirements.  The big caveat with something like that is that you have to proofread it afterward to make sure it still makes sense.  AI is a tool, not an answer. 

There are a lot of AI programs about there.  So, for the next few weeks I am going to look at some specific AI programs, sorted by type. Remember, these are being published in August 2025.  By September, they will be out of date.  Here is my big secret though – I used AI to get this information.  I searched the general internet for this information and could not find anywhere that discussed specific brands by feature. Gemini could and did with the prompt “Which AI programs do what tasks?”  I imagine there are a lot missed, but I did look each one up so I could provide a hyperlink. At the very least, everything I include in these next few discussions actually exists.  Additionally, I removed every program that does not have some level of free use. 

Gemini tells me there are seven basic types of AI:

1. Natural Language Processing (NLP)
2. Computer Vision (CV)
3. Generative AI (Art, Music, Code, etc.)
4. Data Analysis and Prediction
5. Robotics and Automation
6. Healthcare Applications
7. Financial Modeling

Over the next few weeks, we will take an in-depth look at each of these, with a heavy emphasis on programs that will most likely help with schoolwork.  We will start next week with Natural Language Processing, the biggest group.

When Chat GPT broke over the world three years ago, I did not know anything more about AI than any other ordinary person.  I did know that it was important and it could seriously change the world as we know it. I also immediately saw its potential for students to use it to do their homework for them.  That is the primary thing I have been studying and learning since the beginning.  There are a lot of really useful ways for everyone to use AI.  Students have to remember, though, that they are learning and sometimes you just have to things the hard way in order to make them stick. 

One of the places I get my information is a blog called One Useful Thing by Ethan Mollick.  Early this month, he published a blog called Against ‘Brain Damage’ exploring how AI can help or hurt the way we think.  It had a lot of useful information and I may revisit some of the topics later, but today I want to look as a specific study.

According to Mollick researchers “at Penn conducted an experiment at a high school in Turkey where some students were given access to GPT-4 to help with homework.  When they were told to use ChatGPT without guidance or special prompting, they ended up taking a shortcut and getting answers.  So even though students thought they learned a lot from ChatGPT's help, they actually learned less - scoring 17% worse on their final exam (compared to students who didn't use ChatGPT).”

The same study surveyed the students four months later and found students who had used Chat GPT remembered less than students who did not use AI for the same assignment.  What it boils down to is that getting the answers is not the same as finding the answers.  Do the work, do the research.  I am able to pull information from papers I wrote twenty or more years ago out of my memory because I had to learn the material as I was working on it. You can (and should) too!

Did you know that you can search Consensus with a conversational (and maybe controversial) command?  

Try this:  How does immigration impact local economies?  Group together the pros and cons.

The answer was quite a bit longer than I want to include here, but you can easily see for yourself that there are a lot of possibilities with the answer.  Most importantly, it provides the source material.

Your college experience should be about exploring a variety of points of view.  Questions like this can help you see beyond what you have experienced so far.

If you are interested in current, and possibly controversial, events, try exploring Our World in Data, which provides accurate, sourced statistics on almost every topic.  As good as Consensus is, it only provides information from peer-reviewed journals.  Most of the time that is enough, but there are a few websites in the open source world that are worthy as well.

Chain of Thought (CoT) prompting is a technique used to improve reasoning by encouraging models to break down their thought process into intermediate steps.  This method helps the model articulate its reasoning, leading to more accurate and transparent answers.  

An important reason to use chain of thought prompting is that LLMs are good at finding patterns in massive amounts of data, but they often struggle with complex reasoning tasks.  They might give the right answer by chance, but not necessarily because they understand the underlying logic

CoT prompting addresses this by guiding the LLM through the reasoning process step-by-step or using a chain.  This chain typically consists of intermediate steps, justifications, or evidence used to arrive at the final output.  We do this by providing examples that show how to break down a problem and solve it logically.

The user explicitly prompts the model to generate a step-by-step explanation or reasoning process before arriving at a final answer.

Try this example:  Solve this problem step by step. 

Problem: What is the value of 5+7+9-12? 

Response:

To solve the problem 5+7+9−12, we follow the order of operations from left to right:

1. Add 5 and 7:  5+7=12
2. Add the result (12) and 9:  12+9=21
3. Subtract 12 from the result (21):  21−12=9

Therefore, the value of 5+7+9−12 is 9.

The statement ‘solve this problem step by step’ is the most important factor in this process.

You will find this style of prompt creation to be particularly effective for conducting research.  CoT prompting builds on the principle that complex problems are often best tackled by breaking them into smaller, more manageable pieces.  We know this is true in any kind of research, so why not get the computer to help you.

You have to define what you are looking for, so some careful thought will have to go into the prompt.

Example:  Provide three subtopics for a paper on women soldiers’ direct combat contributions in the revolutionary war.

I tested a basic zero-shot prompt, women in war.  Consensus provided two pages of topics, any of which would make several good research papers.  It was a lot to sort through and AI is supposed to reduce your work.  

Next, I tried a CoT with clearly defined and detailed parameters “Three Subtopics for a Paper on Women Soldiers’ Direct Combat Contributions in the Revolutionary War”

• Women Disguised as Soldiers and Direct Battlefield Participation
  • Explore cases like Deborah Sampson, who disguised herself as a man to fight, and other women who directly engaged in combat or took up arms when needed, such as Molly Hays at the Battle of Monmouth.
• Overlapping Roles:  From Camp Followers to Combatants
  • Analyze how women’s roles as camp followers, nurses, and support staff often overlapped with direct combat activities, blurring the lines between support and soldiering in the heat of battle.
• Recognition, Memory, and Historical Marginalization
  • Examine why women’s direct combat contributions have been underrepresented in historical narratives, and how modern scholarship and movements have worked to recover and recognize these stories.

This response provides plenty of examples to start your research on as well as three legitimate sources for each topic. 

Few-shot prompting is a technique in artificial intelligence where a model is given a few examples of a task to improve its performance on similar tasks.  This can be used after a zero-shot prompt to increase specificity about your topic or generated right from the beginning.

For example:

“Regular exercise improves cardiovascular health by strengthening the heart and increasing circulation.  It also enhances mental well-being by reducing stress and boosting mood through endorphin release.  Now, describe another benefit of regular exercise.”

It is also an excellent way to find points of comparison or contrast in two or more items.

Enter the prompt statement and include the items to be analyzed in quotation marks.

Classify the sentiment of the third movie review.  Use the information from the first two examples: 

Review:  "This movie was a waste of time."  Sentiment:  Negative

Review:  "I couldn't stop laughing throughout the film!"  Sentiment: Positive

Review: "The special effects were amazing, but the plot was confusing.  "Sentiment:```?

WARNING:  This example does not work well with Consensus which looks at journal articles to find real data.  It did work well with Google Gemini, a more generic tool, which provided this response:

The sentiment of the third movie review is:  Neutral.

Another example is:

Write a funny limerick about a boy named Sue  This contains two ‘shots’ of information, boy and Sue.

A young boy named Sue, quite unique, 

Had a name that made everyone shriek. 

He'd say with a grin, 

"It's not where I've been, 

But the stories I've got, so to speak!"

You can try again, over and over until you get something you like

A tough boy named Sue, quite perplexed, 

When folks thought his gender was vexed. 

He'd just spit on the ground, 

"That's a man's name, profound! 

Don't you know Johnny Cash? Read the text!"