Generative AI - 6 things to know before falling into the marketing trap

Move on Metaverse, it’s 2023 and Generative AI has already hijacked most social media conversations (and perhaps some boardroom conversations too!). The FOMO around generative AI can be seen with people publishing AI-generated content like essays, poems, art, illustrations, and even jokes. They are either glorifying its capabilities or ringing the death knell on human jobs. But generative AI has been around for quite some time.??

I would know, since I’ve built two startups using AI technology. My edtech startup, SelectQ, created performance-based training quizzes and practice tests fully tailored for every student using AI. At our current startup, MachEye, we have successfully built an analytics platform that understands natural language questions, extracts personalized actionable insights, and presents them as interactive audio-visuals and text narrations generated on-the-fly using AI.??

So, believe me when I say that we need to take this “new” hype of generative AI with a pinch of salt. My experience of applying it for the past 7 years should be enough testament. Through this blog, I want to share 6 things that you should understand before considering generative AI for your business purposes.?

Disclaimer: This is not an AI-generated blog :-)

1. The value of content is in the value of the author?

“It isn’t what we say or think that defines us, it is what we do” These famous words by Jane Austen remind us of the value we bring through our creations. The works of great philosophers, thinkers, authors, and poets are known and quoted worldwide today because of their depth and credibility. On the other hand, summarizing content and rehashing it several ways to generate so-called “new” content brings no real value. More quantity in less time leads to a content tsunami, which ultimately increases mediocrity, dilutes originality, and obscures content that’s actually valuable. For example, I provide generative AI a prompt to “write a poem on war, drone attacks, and evacuation”. I have never experienced war in my life (for which I’m grateful). So, will my AI-generated poem match the authenticity and depth of emotions felt by a war refugee who had to flee and leave everything behind? And what value will it bring to the world? I doubt.?

Instead, can generative AI be used to create real new content that matches a specific business requirement, like product documentation, case studies, request for proposals, or tutorials? Now that’s something useful.

2. Think search is dead? Wait till Bing improves it with generative AI.

“I no longer need to use search”, “Google search is dead” are some arguments I’ve been hearing from generative AI fans. You could just ask a generative AI app a question and it will get a single comprehensive answer, instead of infinite list of search results. Useful to a certain extent, but enough to eliminate search engines? No, not yet. Generative AI needs to be fed a lot of existing data for it to learn and deliver content. Data sets are limited, not current, and take time to be processed. One iteration of training costs almost 10MM. On the other hand, search engines index content continuously, every minute and every day. The information you can get from a search engine today is more current and comprehensive than a generative AI app. I tried asking about the winners of the recently concluded FIFA World Cup 2022 and got the reply “I'm sorry, I do not have knowledge of the latest FIFA World Cup as my training data has a cutoff date of 2021. The most recent World Cup that I am aware of was held in 2018, and it was won by France.”

Forget asking about weather forecasts, latest news and traffic situations! Search engines also show different types of useful and related content along with search results. These include definitions, summaries, tips, related searches, images, videos and news extracts on the searched term.?

For expecting generative AI to behave as search engines, a lot more investment would be required. It will be an effective investment only when the cost of training is accommodated by the profits yielded from using generative AI as search engines. That means that if the cost of training is 10M, and the training has to be done every 2 seconds, the profits from bing search for every 2 seconds needs to exceed 10M (you can do your math to calculate the profits and associated revenues on an yearly basis!).

Even if that happens, will we be content with search results that are still an hour old??

3. Look beyond past use cases for generative AI?

The popular use cases being tried out today with generative AI have always existed in the past. Technology already exists to use keywords and compelling headings to cater to algorithms and ensure better rankings. Yes, generative AI can give you 100 post headings in 3 seconds, but are those 100 really relevant? People are able identify click-bait content and know how to assign credibility to data sources. Yes, generative AI can help you can create art without knowing the design tools or theories. But by applying filters, changing backgrounds, and stereotyping images, an illusion of originality is created and trust is broken.

Yes, generative AI can give you a marketing plan for the full year in 5 seconds, but will it consider the impact of an upcoming World Cup event or a local trade fair on your marketing strategies? Using generative AI for these mediocre purposes seems like a wasted opportunity and leads to increase in spam content. It is like using a high-end DSLR camera lens for your daily selfies!??

Explore forward-looking use cases where the capabilities of generative AI can be put to achieve real business value. For example, to identify hidden bias and cultural appropriation, ensure adherence to company style guides and policies, or convert highly complex jargon-filled language to simpler easy-to-understand language. Or to pacify a client who insists that the brand mascot be made to look 5% happier and no more!?

4. Technology is not mature until it asks a question back?

Asking questions fuels curiosity and shows that you are actively listening and processing the information you are receiving. It also helps clear confusion, reduce ambiguity, and confirm understanding. That’s what a search engine does you enter “today’s whether”. It understands that there’s a confusion in words, checks with you if you want results for “today’s weather” instead, and still leaves the option open to search for the incorrect word. This exchange gives you the confidence that the search engine is not hastily delivering results without considering the context and verifying the request. Generative AI does not ask you questions for clarification. It gives you a universal answer which might not be accurate given the different contexts that can emerge. For example, I asked generative AI how to troubleshoot a Sony TV that turns of randomly. It gives a list of tips, but mind you, those might be applicable to only certain models. If it does not ask for the make or model or year of manufacture, you are missing out on specific answers.

Ambiguity correction and contextualization become very important in generative content. If generative AI keeps giving generic answers, how do you trust it? Will you really be able to solve your business problem with such content? Why does the technology answer irrelevant and incomplete questions in the first place without understanding its relevance or confirming its context??

5. Programming Generative AI does not happen by magic?

Like writers and designers, software developers are also facing concerns about generative AI’s impact on their work. Yes, it can generate code for a full application within a matter of seconds. But software development is not just about writing code. Coding still remains only 10-20% of the overall software development effort. What’s important is getting the requirements right and converting them into solutions that solve problems. It is the collective effort of human thinkers such as business analysts, subject matter experts, developers, testers, user experience designers and product managers who empathize, improve, and continue to enhance solutions.

GitHub’s Co-Pilot is still a better use case of AI, as it provides code suggestions and saves time creating boilerplate and repetitive code patterns. But generative AI falls short of expectations. Instead, can generative AI convert design specs directly to build a requirement-specific application? Can it simplify the task of identifying and fixing issues related to security, privacy, and vulnerability? These capabilities can be of minor help but they won’t be enough to solve bigger problems. So, to my fellow human developers, don’t worry, you will still be wanted!

6. Understanding the user’s intent (or the lack of it!)?

Generative AI mainly imitates and churns out what it is fed. It cannot differentiate between correct, misleading, partial, and biased information. It does not ask for context and delivers answers that are either hits or misses. Till it does not understand the intent behind a user’s request, it will be difficult to trust what it generates. To make it further worse, generative AI is made with a “never say no” mindset. Even if there is no relevant answer, it forces a random construction of content, even if it is irrelevant to the user’s intent. For example, I asked “why only a couple of devices connect to an ASUS AiMesh node?” Now, before starting the troubleshooting, anyone would have confirmed some basic prerequisites like the model number of ASUS device and the backhaul mode in question. But generative AI throws a random list of tips, none of which seem relevant.?

Google acquired DialogFlow 8 years ago as the first AI implementation towards automating customer interactions. However, it turned out to be a non-scalable solution. Getting the same customer interactions practically doesn’t and shouldn’t happen in an enterprise setting frequently. Most AI technologies can help with NLP to show the next course of action like pointing to a help page. In such interactions, NLG is of no use.?

Instead, can generative AI build an FAQs system based on user queries, support tickets, and interactions? Can it serve initial queries of users and then handoff to a human support person? Seems there’s still some time before we can expect bot-based response interactions from generative AI.

Jack of “writing prompts”, Master of none?

Beyond all the current hype, when you examine its capability to solve real business use cases, what generative AI does is very little. It solves unnecessary problems and caters to mediocre wants. It borders on plagiarism by feasting on open source and manually generated code. The basis of open source is to improve transparency, identify vulnerabilities, and contribute towards a collective good. However, it seems wrong and unethical to use public property and invest so much, just to create more spam.

For us humans, writing, drawing, composing music (or for that matter, even creating code!) are fundamentals ways of self-expression. We create content for other humans to share our knowledge and gain new perspectives. Then came the algorithms, and in the race to be on top of the list, we started creating content targeted for web crawlers and bots. Now, with generative AI creating overwhelming amount of content for attracting algorithms, I wonder if we are removing humans out of the equation and what we are gaining by it?