HOW CAPABLE IS ARTIFICIAL INTELLIGENCE?
Who will be the masters of the future?” is an article I published on October 4, 2020 (https://en.muratulker.com/y/how-artificial-is-artificial-intelligence-who-will-be-the-masters-of-the-future/), delving into the history of AI. Written at a time when ChatGPT3 was taking its first steps, the article takes readers on a little journey in history, all the way back to the ancient times. I told about the ten scientists who were interested in machine intelligence who came together for a six-week workshop at Dartmouth College in the US in 1965 and used the term “artificial intelligence” for the first time. I also mentioned Cahit Arf’s speech in 1958 on artificial intelligence in Erzurum, Türkiye.
Yahya Ülker shared very nice post in Turkish on LinkedIn about the history and future of Artificial Intelligence, and I strongly recommend you read it to get an idea about the evolution of AI (https://lnkd.in/dtNBjTff) So, what happened since then? Where do we stand now? Every day, many columns, scientific articles, and books, some of which end up repeating the same thing, are published on the topic. Let’s take a look at the developments together…
In one of my articles from October 4, 2020,”HOW ARTIFICIAL IS ARTIFICIAL INTELLIGENCE? Who will be the Masters of the Future?” (https://en.muratulker.com/y/how-artificial-is-artificial-intelligence-who-will-be-the-masters-of-the-future/), which also delves into the history of AI, I wrote:
“For this reason, it is not artificial intelligence that will create today’s advantage, but the courage and agility of our employees. Some say, “If your employees do not use their potential for creativity, empathy, and problem-solving, then in the future, just assign all your work to AI-driven machines.” If our jobs were repetitive, routine, or predictable, we would compete to automate our work urgently. That an automation revolution has taken place is certainly true. What computers and machines cannot do is to decide how to manage the company and the machines.”
Artificial intelligence models are now called “Large Language Models (LLM)” and the interest of workplace employees, individual performing artists, writers, and designers in how to benefit from Artificial Intelligence (AI) to achieve various goals and perform tasks has rapidly increased its use.
Today, there are dozens of artificial intelligence software rivalling ChatGPT alone, such as Microsoft Bing, Perplexity AI, Google Bard AI, Jasper Chat, Chatsonic, Pi, YouChat, Replica, DailoGPT, and Open AI Playground. If you go a little further, you could add Copy.ai, Grammarly, DALL-E 2, Midjourney, Mem, Fireflies, Slidesgo, Wordtune, and Krisp. You cannot master the list because when you search, an artificial intelligence tool gives you a list that includes itself and aims to persuade you to use it. Of course, there are moral problems brought about by artificial intelligence, such as losing contact with reality or being more easily deceived. I mentioned this in my first article, and I will repeat it at the end of this one.
Let’s take a look at how many days it took for some platforms to reach their first one million users: ChatGPT achieved this in 5 days, Facebook in 10 months, Instagram in 2 months, Spotify in 5 months, and Netflix in 3.5 years. Aren’t these some interesting numbers? Undoubtedly, the network structure of the global community was different when each of them was launched, and we are rapidly evolving into a global society that is connected with more networks as the language barrier disappears day by day (1).
In the past three years, efforts have been made to increase performance with artificial intelligence applications in all sectors, from education and art to production and logistics. Potential applications of artificial intelligence continue to expand as more people embrace digital technologies.
The use of artificial intelligence stands out, especially in finance, digital areas (such as social media, e-commerce, and e-marketing), healthcare, and legal services. With AI-inspired modern art and Elon Musk’s contributions, it’s no surprise that AI continues to dominate the discourse everywhere.
As the scope of artificial intelligence use expands, more ways to improve our personal and business lives will be discovered. It’s a very attractive field: The global artificial intelligence software market is expected to reach USD 22.6 billion by 2025.
As I mentioned in my first article, artificial intelligence is a very complex subject that only software engineers can truly understand, involving many layers. A “true” AI is a machine that can simulate human intelligence, behavior, and even emotions. So, it’s still “artificial”. It’s not real intelligence. Whether artificial intelligence will be equated with real intelligence is a whole other question. In my opinion, it seems unlikely because reading the meaning of transactions and our decisions involve immediate emotions. But who knows? One cannot help but leave an open door in the face of such rapid developments.
Recent developments: The launch of ChatGPT gave the world a glimpse of what future chatbots could look like. ChatGPT engages users by chatting, answering questions and even challenging certain ideas, and it is better at it than I am.
Thousands of companies have adopted using AI-based chatbots to provide 24/7 customer support and resolve issues quickly. As artificial intelligence continues to develop, chatbots’ use of language may evolve and become more complex.
It may be surprising to hear that artificial intelligence applications, popular for their ability to solve problems and make decisions using data input, recognize and interpret visual information, and recognize, interpret, and respond to written and spoken language, are used even in agriculture. Artificial intelligence has created applications that can detect deficiencies in the soil and offer planting recommendations accordingly. In this way, farmers can achieve the precise agricultural planning they need. Additionally, combining artificial intelligence and robotics helps farmers harvest crops faster and more efficiently than they can with human workers.
The e-commerce industry still benefits greatly from artificial intelligence. Companies use AI to predict trends, analyze performance, help manage inventory, and more. AI’s ability to track usage patterns and verify information has also made it a powerful tool in combating credit card fraud and fake online reviews.
Additionally, AI is the basis for “recommendation engines” that scan products for shoppers and show them based on their history and preferences. And this is where virtual assistants and chatbots come into play.
While education is still a human business, artificial intelligence is also helping to expand the potential of educators. Artificial intelligence is often used to facilitate automation in repetitive and data-intensive tasks. Chatbot-style artificial intelligence comes into play again here -this time to answer routine questions quickly, allowing educators to spend more time on complex tasks.
The use of artificial intelligence in education to close the education gap in Turkey deserves to be addressed in an entire article on its own.
The field of finance has largely transitioned to using artificial intelligence at all levels. Customers use artificial intelligence to get information about their banking and investment accounts. Banks and credit card companies rely on AI to detect changes in transaction patterns to catch fraud in the act. Creditors use artificial intelligence to predict and evaluate borrowers’ risk levels in making loan decisions. Venture capital firms embrace AI to generate customized insights and financial risk management decisions.
Artificial intelligence is also taking center stage in the medical field. Artificial intelligence allows hospital administrators to process data, schedule meetings, organize files, and helps them transcribe medical histories. With the help of artificial intelligence, robotic machine-led surgeries are more precise, have a smaller margin of error, and can be performed 24/7. There are many examples of artificial intelligence surpassing humans in diagnosis. Pharmaceutical companies use AI to analyze historical and modern data to discover new potential drugs.
Another common application of artificial intelligence is in marketing. Artificial intelligence’s ability to analyze data and quickly create insights is very useful for teams that must act on them. Artificial intelligence serves to create campaign reports, improve customer engagement, personalize messages, optimize campaigns, and even change ads during the campaign based on new analysis. Brand messages can now be produced instantly, with personalized content.
Companies like Meta, X, and Google are using AI to analyze large amounts of data and create actionable insights:
They track user behavior to create marketing and advertising tactics. They monitor comments to suggest new posts and accounts to follow. They set trends. They help create content based on demographic and behavioral data. They can also combat cyberbullying and harmful or illegal content.
Artificial intelligence is also a key enabler in autonomous driving programs used by Tesla, Audi, Volvo, and others.
Our smart digital devices also use artificial intelligence to log in and authenticate with facial recognition programs.
Artificial intelligence now also exists in “in-home” applications.
Devices that we can now call home robots, such as automatic vacuum cleaners and lawnmowers, rely on artificial intelligence to avoid obstacles, learn, and work efficiently.
And, of course, various advanced home security systems, as well as Siri, Amazon Alexa, and Google Assistance, which are not very commonly used in Turkey, also rely on artificial intelligence. Whether you realize it or not, your e-mail account is probably using artificial intelligence to filter out spam and illegal content (1).
Yahya Ülker’s LinkedIn post, which I mentioned in the introduction, points to where this could lead very well: If these developments continue, we will witness the birth of artificial superintelligence in 30 years. Integration of humans with AI may occur within 40 years. Does a future with robots with human-level consciousness await us in 50 years?
For example, in 2030, realistic videos created by artificial intelligence in the entertainment industry and music, surpassing the talent of famous musicians will exceed the limits of our imagination. At the same time, artificial intelligence language models will be able to have in-depth conversations with all people. Prostheses to heal the blind or enable paralyzed people to walk may also be developed, becoming an ordinary part of our lives.
By 2040, “artificial general intelligence” may emerge thanks to detailed scans of the human brain and trillions of virtual simulations. Artificial intelligence, which would have more abilities than the average person, would be connected to the worldwide internet network and capable of building new, improved, and more advanced versions of itself.
By 2050, “artificial superintelligence may reach a level that will surpass human intelligence to a significant extent worldwide -becoming, perhaps, a billion times better. It could reach a level where it makes discoveries worthy of a Nobel Prize in seconds and easily makes inventions that would take centuries for humans to make. There could be advanced humanoid robots that can think and act like humans (2).”
Ultimately, what is certain is that artificial intelligence is transforming all sectors. Not a day goes by when we don’t hear some news or read about a new development or application driven by artificial intelligence. Advanced artificial intelligence models such as ChatGPT4 are already being used to optimize and improve resources while creating many products and services. So, the question is no longer whether there will be artificial intelligence or not, or whether we should use it or not. Now, we must come up with ways to leverage such artificial intelligence models or programs efficiently. To this end, we as the users must use “prompt engineering,” which could be described as blowing (communicating) the right message into the ear of artificial intelligence. Because, you know, trouble shared is a trouble halved!
Prompt engineering and communication with artificial intelligence is described as the science, and even the art, of creating effective input infrastructure (3).
In artificial intelligence models, significantly large amounts of data are stored in certain relationships within complex language models. Thus, these tools perform various NLP tasks, such as machine translation, summarization, and conversational speech.
There are steps in the communication journey of a prompt (command) within artificial intelligence software. In GPT models, for example, there are words or characters called “Tokens.” In other words, the data is expressed in different values, which are then converted into software involving various layers, namely Tokenization. If your commands are not formed in a way that takes the software architecture into account, you will not get the result you want. Let’s give a communication example:
Me: How many eyes does a giraffe have?
Chat GPT: It has two eyes, like other mammals do.
Me: How many eyes does my foot have?
Chat GPT: Your feet have no eyes. Eyes are located on the heads of animals but not on the hands or feet.
Me: How many eyes does my “foot” have?
Chat GPT: I think my foot has another meaning here. Within this context, I can say that your feet do not have eyes.
As you can see, I gave the artificial intelligence a prompt using only the “…” sign, ensuring it understood the context correctly. This was a relatively easy example. The correct selection of the context window is very important in many areas:
–Context Window Selection: The given prompt should not exceed the token limit (for example, this limit is 4096 for GPT 3,), and if necessary, should be divided into smaller pieces, thus reducing the risk of missing the context.
-Artificial intelligence modelling includes several stages, including embedding, probability distribution, token sampling, coding, and decoding, which software developers are aware of and that prompters should at least be aware of, and there is a tremendous probability distribution modelling working in the background. The machine reviews the probability distributions within milliseconds, trying to provide you with the right option. In a sense, when the artificial intelligence suggests ‘chicken translate’ for the translation of the well-known Turkish dish ‘chicken doner’ becauase you did not temper it properly, you need to be able to predict whether such a possibility would come your way or when it does, when someone makes this joke, you should be capable of concluding someone did not quite understand the context.
Prompt Engineering rests on four principles: 1) Being clear and precise, 2) Presenting contextual information correctly, 3) Being familiar with the format used, and 4) Cleansing the prompt off of the crowd of words. Applying these principles affects the performance of artificial intelligence models and boosts the application’s success.
What I have told so far may seem to you a bit like the “mandatory moves” in ice skating. And beyond that, there are the artistic moves of prompt engineering. Sensing how the AI will react to different prompts, questions, and expressions, using certain words strategically, or constructing sentences in a certain way affects the AI’s performance. Step by step fine-tuning is another aspect of the art of prompt engineering. The answers given by the AI need to be examined at every stage and directed at the right stage without thinking, “Oh, this intelligence is talking nonsense.” Meanwhile, it is also possible to shape the tone, style, and level of detail of the prompt according to the user’s preferences. Another important issue is that prompt engineers benefit from linguists and psychologists on issues requiring expertise and other relevant experts for cultural nuances.
The field of artificial intelligence is developing very quickly, with a new model emerging every day, and for each new model, prompt engineers need to review all the elements mentioned above to adapt them to the prompts they provide. It seems more attention is devoted to how to give instructions and prompts to the artificial intelligence model in the relevant field today rather than in which areas artificial intelligence could make things easier for us. However, some experts say that future artificial intelligence models will be more useful and not require prompt engineering (4).
Why is Artificial Intelligence very important in our business lives? Now, let’s return to our main topic and see how we can benefit from artificial intelligence in company management.
Artificial intelligence brings the right data at every level, from production and sales to senior management and employees, at just the right time to support their decisions. However, artificial intelligence cannot contribute much to these companies if internal bureaucracy is so inricate as to prolong decision-making. Companies wishing to benefit from artificial intelligence should most particularly democratize their decision-making mechanisms (5).
In their Harvard Business Review article published on the January-February 2023 issue, Davenport and Mittal examined 30 companies that successfully implemented Artificial Intelligence applications and identified the ten basic actions they took. You may find a list of these ten basic actions below (6):
1. Decide What You Will Achieve With Artificial Intelligence Tools
2. Work with a Partner Ecosystem
3. Specialize in Analytics, That is, Data-Driven Decision Making.
4. Create a Modular and Flexible IT Architecture
5. Adapt AI to Existing Workflows
6. Develop Solutions to Suit the Entire Organization
7. Create an AI Governance and Leadership Structure
8. Create an Employee Center of Excellence
9. Invest Constantly
10. Research New Data Sources
I agree with all of these steps, but I would highlight the 7th. A good leader generally knows what can be done with artificial intelligence tools in their business. They can see how it can be implemented with the strategy, business model, processes, and people. What’s important here is to create a decision-making culture which would enable the utilization of artificial intelligence. Artificial intelligence does not make decisions on its own.
It is no secret that decision-making based on productive or creative artificial intelligence models will be the center of competitive advantage in the short term. It does not seem possible to increase the decision-making abilities within a company simply by providing employees with the ability to benefit from artificial intelligence or by providing them with prompt engineering skills. Companies need to radically update their decision-making systems as required by the changes in artificial intelligence applications, that is, democratize and scale artificial intelligence into the entire company’s decision system, and use it as a leverage point!
AI-driven decision-making is a broader concept based on human-machine collaboration. When the moment of decision comes, it is not just about looking at the data and coming to a conclusion. It is rather a process that requires more extensive evaluation, asking the right question and framing the problem correctly. The important thing here is empowering more employees to make free decisions!
The C-Level’s job is to supervise how this free decision-making process is used. Artificial Intelligence should be employed along with a certain standard of morality to ensure its use is fair, safe, and sustainable, and prevent Artificial Intelligence models from creating inaccurate information or answers that contradict the company’s values. Of course, employees need to be trained about the proper human-machine decision-making concept (7).
Recently, a group of Harvard researchers conducted a field experiment with 759 employees constituting 7 percent of all the consultants at the global management firm Boston Consulting Group, to investigate whether ChatGPT increased the quality and performance of more talented employees and published the results (8).
After establishing a performance baseline for a similar task, subjects were randomly tested in one of three scenarios: no access to AI, access to GPT4 AI, or access to GPT4 AI with a prompt engineering overview.
The researchers assumed that while artificial intelligence makes it easier for employees to complete some tasks, creating “pioneers who are better at using technology.” Still, some situations and tasks exceed the capacity of artificial intelligence even if they appear to be of the same difficulty level.
For 18 real consulting task sets, consultants who used AI on tasks that were within the limits of AI capabilities were significantly more productive, that is, by 12.2 percent on average and completed the task 25 percent faster and with better quality, i.e., 40 percent higher than the control group. Advisors with different skills benefited significantly from the support of AI, meaning that the average performance thresholds of those above the average increased by 43 percent whereas the performance of those below the average increased by 17 percent (8).
This study shows us that human-machine communication needs to be monitored closely indeed, that different results could be obtained depending on tasks and individuals, and even that with the addition of different artificial intelligence models, quite different results could be obtained. In other words, implementing Artificial Intelligence in companies requires a whole other attitude than, “Let’s install the program on the computer, then give it some training so that everyone can do it,” that is, an analytical thinking system, and a completely different management approach. Today, companies seek artificial intelligence experience in the job applications they receive but as I emphasized above, what matters here is if they have the aforementioned decision-making environment to begin with? (9).
The most controversial aspect of Artificial Intelligence is that the software is not transparent, with such technologies as deep fake making it easier to mislead and deceive people, and t the information produced in this way inflicting serious damage on intellectual and copyright rights. Of course, there is also the concern about the formidable control basis it creates for the states to keep society under control (10).
About six months ago, the Future of Life Institute, a non-profit organization, shared an open letter signed by many famous people, including Elon Musk, Steve Wozniak, and Yoshua Bengio. The letter called on tech companies to “pause” the development of AI language models more powerful than OpenAI’s GPT4 for six months. Of course, nothing such thing happened. In an interview, the President of the Institute said, “In the meetings we held to create this letter, I saw that the giants of artificial intelligence are aware of the risks, but no one dares to speak. This letter couldn’t stop the work, but at least encouraged people to speak up.” How Artificial Intelligence will be regulated is still a matter of debate. Some want to establish an authority with institutions such as RTÜK (the Turkish Supreme Board of Radio and Television) and the FDA. Some want it to be governed by global institutions, such as the United Nations and NATO, while others call for laws and regulations to be enacted urgently and for this business to be subject to state licensing (11).
On the other hand, it is also said that some employees and even some businesses are afraid of Artificial Intelligence because they believe they might end up losing their jobs or superiority in the industry (12). Of course, the solution is to learn, gain skills, and not lag behind.
Already, people are talking of interactive artificial intelligence, where you can give prompts by speaking rather than writing, and it is said that this artificial intelligence will give contextually better results than giving prompts by text. One way or another, this technological development has already launched and it is not something that can just be given up on. To prepare your companies for artificial intelligence transformation, I recommend you read Alexander Borek and Nadine Prill’s book, “Digital Transformation with Data and Artificial Intelligence” (13). I believe this book offers the essentials of business and must be read by managers at every level in the entire organization. Let me end my article with a quote from the foreword I wrote to this book:
“The fluctuations in digitalization and artificial intelligence follow an exponential curve, meaning that they come slowly, unnoticed, and then suddenly, as in the case of Nokia and Kodak. They ruthlessly disrupt your entire operation. The digital maturity level of traditional companies is still extremely low. Because when the life cycle of their products is at the ‘milking cow’ stage, revenue and profit also stand high for now. Thus, senior managers do not yet feel the digital fluctuation in their sector. Even if they believe it will eventually happen, it takes many years to see the first results once it does. There aren’t many managers in the organization who remain in the same position for that long, as many either move on to new jobs or retire by then. Why would someone risk their current position in the organization for something that won’t affect them immediately? Only if the current CEO correctly properly drives this potential change in digitalization and artificial intelligence employees will adapt and acquire new skills. What boards responsible for driving digital transformation, especially in traditional manufacturing businesses, do wrong is they believe digital transformation will magically happen in their companies if they appoint a top manager responsible for data. However, most keep managing their businesses based on the core beliefs they have adopted for years. Changing these beliefs is impossible unless the CEO believes in this and takes risks by leading the change. At the very beginning of digital transformation, CEOs and the senior manager responsible for data should take on the role of the chief executive and together demand the necessary changes in the application based on the results obtained. It is only by leveraging the core competencies acquired in this way that companies can gain the following skills and survive in their competitive environment:
1. Collecting, processing, and utilizing high-volume data,
2. Building data science and machine learning software products in the cloud,
3. Redesigning the current business model and adapting the changes.”
References
(1) (2023) Applications of Artificial Intelligence across Various Industries; Forbes, June 6.
(2) https://www.linkedin.com/posts/yulker_30-y%C4%B1l-i%C3%A7inde-yapay-s%C3%BCper-zekan%C4%B1n-do%C4%9Fu%C5 %9Fu-activity-7115229050171719680-VORO?utm_source=share&utm_medium=member_desktop)
(3) Lo, S. L (2023) Art and Science of Prompt Engineering: A New Literacy in the Information Age, Internet Reference Services, June. Routledge.
(4) AI Prompt Engineering Isn’t the Future, HBR. June 6, 2023, audio article.
(5) Di Fiore, A. (2018). Why AI Will Shift Decision Making from the C-Suite to the Front Line, August 3, HBR.
(6) Davenport H. T. and Mittal N. (2023). Stop Tinkering with AI, HBR, January-February 2023.
Note: This open-source article can be quoted by citing the author. Does not require copyright.