Home » INSIGHTS » The AI Training We Really Need
The AI Training We Really Need
We ALL need AI training. Not the training that tells you how to program or use AI. No, we need the training that helps us distinguish AI-created content from human-created content. No matter what it is — news, images, music, art, articles, etc. — wouldn’t you like to know who is serving you the images and videos you consume daily? To learn more about distinguishing AI and human content, I first went to Google and found a great article on the Tidio blog. Tidio sells chatbots and other AI tools, making them well-schooled in artificial intelligence. They created an AI-test to see how well we humans could identify AI text and images. A few unfortunate data points emerged from their study:
- We’re better at identifying AI-created cats than humans.
- AI images created in 2018 were mistaken for real by only 35.7% of those tested, but the images generated in 2021 fooled 68.3% of us.
In other words, as predicted, AI is getting better and better, making us more and more prone to mistaking AI content for real.
Distinguishing Human from AI Content
Ironically, when exploring AI Training, I asked ChatGPT for some tips on differentiating AI and real content. I learned to look for:
- Unusual or inconsistent behavior: AI-generated content may exhibit patterns or behaviors that seem slightly off or inconsistent. This could be due to the limitations or biases in the training data or the specific AI model used.
- Rapid response or superhuman output: AI models can generate an extensive amount of content at an incredibly fast pace. If you notice an immediate, high-volume response, it could be an indication of AI involvement.
- Lack of emotional or personal context: AI-generated content often lacks personal experiences, emotions, or subjective perspectives. While it can mimic human-like responses, it may struggle to provide specific details or real-life anecdotes.
- Specific language cues: Some AI models, like GPT-3.5, offer disclaimers or identify themselves as AI. Some content will explicitly state, “generated by an AI system.” That’s a clear giveaway. However, not all AI-generated content includes such disclaimers.
- Contextual inconsistencies: AI models may struggle to maintain coherence when engaging in extended conversations or handling complex topic transitions. If you observe abrupt changes in subject matter or inconsistency in the narrative, it might indicate AI involvement.
- Uncommon errors or lack of common knowledge: AI-generated content can sometimes make unusual errors or provide incorrect information that a human would likely know. AI also struggles with current events, particularly outdated training data.
The copy is pretty good, right? Still, for me, looking at an image to determine its reality is like playing those spot-the-differences photo games in magazines, where you see a photo that’s been altered and you have to find the 10 changes. Some of those are really hard… and it’s a game you decide to play. The trouble with AI is that you don’t know you’re playing a game. Unfortunately, as visual mammals, we consume so many images and messages every day that we typically breeze through them quickly, not looking for inconsistencies or awkward syntax in writing. We’re bound to be duped!
AI Training to Help Trainers Make Best Use of AI
- Use ChatGPT for ideas, not truths – If you’re using an AI website, put your search results through your own accuracy filter. Even then, be wary: so many articles written in the past regurgitate ideas and information that may or may not be well-researched or correct.
- Get assistance with writing blogs, essays, and copy – A ChatGPT search can help streamline the writing or creation process, but writers will need to add nuance and perspective, in order to differentiate their articles from those previously written. I admit, I frequently poo-pooed articles written just for SEO optimization, which simply reframe existing articles with new, self-serving links. In truth, search algorithms rewarded this approach. Perhaps the ready availability of ChatGPT content will necessitate more creative writing. One can hope!
- Try AI-powered apps to accelerate your development process – Perhaps you’ll use ChatGPT as an idea generator, rather than a source of truth. Don’t build a training session based on a single AI search. Rather, utilize the breadth of reliable resources. Or, use AI-powered video, graphics, and presentation builders to improve your existing content and possibly get you 90% of the way to a “sexier” presentation. These tools might help you build more engagement in the learning process, provide feedback to learners, develop assessment questions, or track performance.
- Remember that customer service chatBots have limitations – perhaps you’ve already experienced the frustration of asking a chatbot a question and getting unhelpful or incorrect answers. At this point, complex questions cannot be answered through chatbots, but they’re likely to improve over time. Because you may not be able to rely on chatbot answers, be ready to send an email or pick up a phone, to speak to a human who can understand nuance and challenging situations.
- Assist with rapid onboarding – onboarding chatbots present the same challenges as customer service chatbots. For FAQs, a bot might be able to provide employees with quick answers, but not understand more complex inquiries. Chatbots can “assist” with onboarding, but they can’t do the whole job. Even playfully- and thoughtfully-programmed bots will not communicate or replace person-to-person emotions, connections, relationships, or corporate culture.
- Draw on Josh Cavelier’s 150 prompts. I just tried this on a topic I know well and found it quite helpful. Cavelier assembled a menu of 150 prompt templates–phrases to write into ChatGPT to yield information needed by instructional designers. For instance, the Course Outline prompt might look something like this: “Act like an instructional designer and create a detailed course outline for an [COURSE_TOPIC] aimed at [TARGET_AUDIENCE] that covers the following learning objectives: [LEARNING_OBJECTIVES]. Include each module’s…” Within those 150 prompts, you’ll find suggestions to help generate useful content for feedback forms, group projects, time management, and assessments, just to name a few.
What is “Real”? A few warnings…
We’ve become so accustomed to photoshopped images and spell-checked documents, we don’t even consider that a cover model might not look as presented, or that colors, room decor, and people in particular backgrounds might not really exist. Nonetheless, we accept them to be art, memes, acceptable advertising, or representations of possibility. Consider, for instance, all the places where Bernie Sanders, bundled up and seated, appeared. At least with memes, we know them for what they are.
Many will use AI apps to save considerable time and money. For the most part, the prevalence AI-created media won’t bother consumers. Within this context, we will remember the caveats:
- Anything you put into the model becomes part of the internet. If you wouldn’t post it publicly, don’t put it into ChatGPT.
- Those with nefarious intent, to spread falsehoods, misinformation, conspiracy theories, and lies have very powerful tools at their disposal.
- AI will tend to whitewash everything and favor centrist perspectives. This can lead us to be colorblind, reinforcing biases in our existing systems and beliefs.
AI Training: Making the Most of AI
As Artificial Intelligence, ChatGPT, and bots are here to stay, the onus is on us to consume it wisely, regulate it as needed, and develop new laws and practices to ensure it becomes a force of good.
Read More
AI Powered Learning Apps – Online Learning
1,516