Human Bookends for AI?

What is the best way to embrace AI?

Wherever you may be on the?AI Worldview Spectrum , if you are looking to use AI in business, you should consider how to leverage AI to its maximum benefit.

While many people debate the distinction between computer intelligence and human intelligence, this perspective fails to approach the real advantage of this fast-emerging technology.

It is not about human intelligence or computer intelligence but rather what my MIT Thomas Malone refers to as “collective intelligence.” The promise of AI’s full capacity is a function of the intersection between human intelligence and machine intelligence. The combined intelligence will certainly outperform the ability of either on their own.

Wharton Professor, Ethan Mollick 's new book,?Co-Intelligence: Living and Working with AI ,?urges us to “engage with AI as co-worker, co-teacher, and coach.”??

Mollick’s book is said to show “what it means to think and work together with?smart machines, and why we must master that skill.”

While many are experimenting with various AI tools, I would argue most are doing so without much direction or training on?how?to best work with them. This is not to disparage these efforts, as most consumers have only access to these tools relatively recently – and they did not come with any instruction manuals.

From my perspective, it is the power of collective intelligence or co-intelligence that will determine the real impact of the future of business (and life, in general).

As a practitioner, and for anyone who's thinking of trying to start using AI, I believe you need to understand that collective intelligence and develop a framework for approaching it.

I offer my framework of Human Bookends for AI.

Whatever project or task you are starting to engage with, it is important not to be a passive participant. Again, from Mollick’s book Co-Intelligence, in the section on his Four Rules for Co-Intelligence, follow “Principle 2: Be the human in the loop.” He states “For now, AI works best with human help, and you want to be that helpful human.”

You need to be active from the beginning through the end. AI can crunch an incredible volume of data, synthesize it, and generate results. The quality of these results, however, will depend on the human bookends on each side of the process.

Initially, it is important to have a human engaged at the start of the project to enter prompts into a Generative-AI large-language model, aka LLM, such as ChatGPT, Gemini, Claude, etc.

This process has become known as prompt engineering. Prompt engineering involves crafting the right prompts or text-based inputs to get the desired output from AI models. The importance of the prompt engineer role was highlighted last year in a?Wall Street Journal article , sharing the emergence of new job listings for prompt engineers with annual salaries ranging from $100,000 to over $200,000. That’s how valuable the role of prompting AI is.??

I am not convinced that prompt engineers will stay the hottest job. Instead, I believe (as do many others) that prompt engineering will become one of the hottest skills a candidate needs to give themselves an advantage in a competitive job market.

The quality of the output from any AI platform will be a direct result of the quality of how you prompt it.??

A couple of overall tips:

· BE CLEAR & SPECIFIC

· PROVIDE CONTEXT

· PROVIDE TONE/STYLE

· EXPRESS DESIRED OUTCOME

All these things are critical to being able to ensure that you're getting as good an answer to your prompt as you can.??

There are several formulas or acronyms people have created to assist in the prompting process. I propose the acronym ARTIF (short for artificial).

A.R.T.I.F.

The A stands for Ask Me. You ask the AI to ask you questions. The answers you provide help the platform to carry out the task. Generally, these are better used for larger projects than simple requests.?

Here’s a sample prompt for someone in HR looking to develop an onboarding strategy:

I am looking to create an onboarding strategy for a new tech startup. Please can you ask me a series of questions, one at a time, to gain comprehensive information for you to assist in creating a robust onboarding strategy document.?

By asking you questions, one at a time, it will understand more about your organization and its needs when it comes to onboarding. It might ask about the size of your business, the specific industry, business objectives, etc. The more context the platform has, the better tool it becomes.

The A isn't always necessary, though when employed it can be very powerful.

I’ll go through the rest of the elements and then after share a comparison of a basic (bad) prompt and a more advanced prompt.??

The R stands for the role that the AI should play. You want to think of the AI as a specific position or as a skilled professional. From Mollick’s Co-Intelligence:

Principle 3: Treat AI like a person (but tell it what kind of person it is).

For example, you could prompt the AI to act as if it were a 20-year veteran of the insurance industry with expertise in life insurance term policies if that would be relevant to the task. Or you may ask it to act as an experienced human resource professional if that’s relevant. Think of who you would wish you had access to and then describe that person. By clearly identifying what role you want the AI to play when it's engaged in the task, it will be both faster and more effective.??

As Mollick shares “The key is to give the LLM some guidance and direction on how to generate outputs that match your expectations and needs, to put it in the right ‘headspace’ to give you interesting and unique answers.”

Prompting it to play a particular role will impact its ability to answer a query with more context and hence will result in improved outcomes.

T stands for task.??

Many prompts simply start with this element, which is fundamental. What are you asking to be done? Again, clear and specific is best. Understand your objective, whether it is brainstorming a list of names for a new company, product, or service in the healthcare space; writing a blog article on the benefits of sunscreen; or adding a particular piece of code for game development. The more you ask in a detailed way the better.

The I is for input. You can upload images, videos, audio, or documents to the platform. Depending on the platform’s ability to handle a particular format. Again, this is more context for the AI. Most of the major platforms are looking to be multi-modal LLMs, meaning they can handle various input formats.

Input can be used in two distinct ways.?

It can be used to provide examples of what you are looking for the AI to generate. For example, if you upload several Picasso images to indicate the style you would like it to create an image in (although it probably already knows Picasso, though you get my point). Or you might provide several examples of shades of purple that you want the image to replicate.

The other way input may be used is in conjunction with the task itself. You might ask AI to summarize a document, rewrite an email, etc. AI platforms are very good at summarization and often help as a writer’s assistant. In these cases, you would want to copy and paste the text into the prompt box, or you can upload the document as an attachment.

Finally, the F is for format. What format do you want the outcome in? Are you asking it to provide an image or video or animated gif? If it’s text-based, is a Word document sufficient? Perhaps you want it to be in tabular format.

I often ask for a tabular format giving the specifics of the columns to create and then it generally outputs it as a .csv file.??

If you ask for a certain format, the platform will attempt to deliver in that format.

Putting all of this together, a simple version would look something like this:??

[ASK] questions to help with [TASK]; Act as a [ROLE], create a [TASK], consider [INPUT], show as [FORMAT].

Now let’s look at a comparison of a basic prompt and a more advanced prompt for the same task. Let’s suppose you wanted to create some ad copy.

Basic Prompt

A basic prompt might be as simple as?“Write me some ad copy for a sports car.”?While it will generate a decent response, it won’t be as detailed or useful compared to a more advanced prompt.

Advanced Prompt

Here’s an example of implementing ARTIF into a more detailed prompt.

As an experienced 20-year copywriting veteran [ROLE] with expertise in writing ad copy for the automotive industry [more context of the role], create dynamic and persuasive ad copy for a new, luxury, electric sports car from Curras [name of brand], the Purple Blaze (sounds awesome, right?). [Task]

Highlight its incredible high-tech capabilities, fuel efficiency, as well as its blazing-fast acceleration that makes you feel like you are a NASCAR driver.

Emphasize its amazing, modern, sleek design and how it will be the best-looking car on the street. It will be so eye-catching, that people will stop and try and catch a glimpse of who is driving this automotive wonder.

The advanced prompt is much longer, much more specific, and more detailed and will produce much more compelling results than the basic prompt.

People often think that a prompt is short, maybe one sentence. That is not the way to generate the best output.

There is a common phrase that has been used to sum up this idea – Better input = better output.

Using AI effectively does not necessarily mean less work. To be used successfully, the human bookend at the start is necessary to properly prompt the AI model, so that it will generate better work.

Other Prompting Tips

Another approach to learning the art of prompting is to ask the model to provide a prompt for any task you are looking to accomplish. If you study the prompts the model suggests, this may assist in future prompting. You can also, in most cases, look at the actual prompt the AI model used. With many text-to-image platforms, it takes your prompt and rewords it. You may not immediately see this revised prompt, but if request the prompt used, or in some cases, there is a built-in way to check what the prompt used was.

The art of prompting is not meant to be a single interaction. As AI evolves (and when AI Agents become commonplace) the more conversational they become. Therefore, you don’t want a one-sided conversation. Once you receive that first response, clarify what more you want. Perhaps it wasn’t formal enough. Prompt it to be more formal. Or perhaps you want it to include something it didn’t initially - you can add this in a follow-up prompt.

Again, in Mollick's book, he shares “One very effective strategy that emerged from class was treating the AI as a coeditor, engaging in a back-and-forth, conversational process.”

This concept was taken to the extreme last year with the “Make It More…” trend with image generation. It has some examples of cute baby unicorn’s getting progressively cuter. It leads to some outrageous results.

All that being said, the concept is sound.?

Continuing the prompt conversation can lead to “more.”?

Another recommendation is that you consider curating a prompt library. While there are a lot of different entities that provide them, including the big foundation models, they tend to be very overwhelming. We are working on one ourselves (watch this space). In the meantime, create your own as you start to use AI, and then pick and choose from prompt libraries that suit your use cases best.

If you prompt effectively AI will generate some pretty amazing output. It would be nice if you could simply take that output and immediately use it. Unfortunately, we are not at that stage of AI development as the tools available simply cannot be relied on without human involvement. I don’t believe there will ever be a time when a human shouldn’t be engaged in the AI process. If you were to simply cut and paste the results, there is a high likelihood that at some stage it will end up placing you in a precarious position.

We need a human bookend at the other end of the process for at least three important reasons.??

At this moment, AI tries to be too helpful sometimes. No matter what you ask it, it will try and answer – even if it isn’t sure about the accuracy of that answer. While all the AI models are working on the issue, at the time of this writing, AI still has what is referred to as hallucinations. It will just make stuff up.

My father, who's an ex-judge, loves the story or loves the example of the story of Michael Cohen, Trump's lawyer. During a lawsuit against Mr. Cohen, his lawyer decided to get some assistance from AI in developing the legal briefs and arguments. AI did its duty and provided some compelling arguments in support of Mr. Cohen’s position. Cohen’s lawyer, however, didn’t check, or check carefully enough the validity of AI’s output. AI provided cases reinforcing the legitimacy of their arguments. The only problem – the cases didn’t exist. AI made them up. AI cannot be relied upon to ensure accurate information.??

When it comes to the human bookend at the end of the process, you must remember: Verify, verify, verify.

While hallucinations will most certainly lessen as AI models improve, it is critical at this stage to ensure someone is reviewing the information.

In addition to verifying that the information is factual, a human actor should also review to ensure that the output provided is accurate for the objective of the task. While the output may not be an actual hallucination, if it doesn’t provide the relevant issue you are looking for, it is fairly useless. A human editor needs to check that the output serves the purpose intended by interacting with AI in the first place.

The final reason you need a human editor at the end of the process is because even with a really good prompt it may end up not sounding on brand. You need to ensure that the brand voice, tone, the personality fit your organization. Hopefully, this can be addressed at the earlier stage when revising the prompt to use a particular voice. Even so, your content must be consistent with your brand voice.

The need for sharing your unique brand voice is such a fundamental element of your message, you cannot afford to forget to ensure this is true of the output of AI.

Like with hallucinations, over time, AI will learn your brand voice. In some platforms, efforts are already underway to strengthen this ability. It may be well written, and it may be thorough, but if it doesn't have the same personality as your other communications, as your other messaging, it won't be as effective.

Overall, AI is an incredibly powerful tool that we could all use right away, but we should be thinking about how we can use it successfully.

By using collective intelligence rather than simple computer intelligence and by having human bookends in place for any AI process or any AI task we can take full advantage of the benefits AI affords.

Interested in how to leverage AI for your business?? Or, looking to create compelling content for your marketing??Contact Geoffrey at [email protected] or call 215.667.1298