The TRUTH about A.I. - Ready in 2025, or Still a Toy?

The Truth About A.I.

A.I., or artificial intelligence, is basically, well, NOT yet intelligent. Think of it like a 10 year-old child, with zero real world experience, who can regurgitate some not-so-nice words she overheard daddy say in an argument with mommy. It’s really that basic at this point, but by the time you watch this, I’m sure it will have progressed.

Still, where the heck are all the robots? Back in the 1980s, we were promised flying cars and home robots by 2020. That turned out to be a huge fail.

And NO, this is not a fake A.I. generated presentation created in some warehouse in rural Bangladesh. And A.I. did NOT write the script. But I did use ChatGPT and Google’s Gemini to help me complete the research for this video, Gamma.app to create the slide deck, and ElevenLabs.io to record the audio using a clone of my voice – a few things A.I. is currently useful for. But, more on that later.

I did a video on the History of Computers about six or seven years back when I was still a computer teacher that scored over a million views on several video sharing platforms. I think I said somewhere in that video that A.I. was still at least ten years from being able to write its own code. I was wrong. It was more like 15 to 20 years. Despite what you think, you still cannot tell any GPT to write a complete FPS game from scratch featuring you versus your favorite Hollywood film stars. A.I. is not yet “intelligent,” or ready for anything beyond basic data analysis and rudimentary searches, and probably won’t be ready for at least another 5 years.

With few exceptions, today’s A.I. is basically an enhanced search engine with some clever pre-programmed analytical tweaks generated largely by clever but poorly paid engineers in nondescript warehouses somewhere in Asia and Africa.

At the New York Times DealBook Summit in November 2023, NVIDIA’s CEO Jensen Huang made the prediction that artificial general intelligence, (A.G.I.), which is just about human-level intelligence, could be achieved within five years. He then made that exact same claim again at the Stanford Economic Forum in early March 2024, and probably a few times after that, so I’m thinking that 5 year timeline is probably a moving target. Is A.G.I.’s debut in late 2028 a reality? Possibly. But I have my doubts. As of the making of this video, A.I. still can’t spell the words you’ve typed into the prompt on images. I’ve been complaining about this for a year and it still has not improved.

How is A.I. being developed?

A.I. development is a collaborative effort involving many different roles and expertise. Here’s a breakdown of the key players, according to Google’s Gemini A.I.:

1. A.I. researchers are scientists and academics pushing the boundaries of A.I., exploring new approaches to machine learning, deep learning, natural language processing, and other A.I. subfields. They often work in universities, research labs, and tech companies.

2. Data Scientists collect, clean, and prepare data. A.I. systems rely on massive datasets to learn. Data scientists are responsible for gathering, cleaning, and preparing this data, ensuring its quality and relevance for training A.I. models. They also analyze the results of A.I. experiments and interpret the output of A.I. models, providing valuable insights.

3. Machine Learning Engineers take the algorithms and techniques developed by researchers and apply them to real-world problems. They build, train, and deploy A.I. models, ensuring they are efficient, scalable, and reliable.

4. Software Engineers create the software infrastructure and applications that utilize A.I. models. They build user interfaces, integrate A.I. into existing systems, and ensure the smooth functioning of A.I.-powered products.

5. Domain Experts provide specialized knowledge. Since A.I. is being applied to various fields, from healthcare to finance to manufacturing, domain experts provide their specialized knowledge and insights to ensure A.I. models are relevant and effective in their respective fields.

6. Tech companies like Google, Microsoft, Amazon, and Meta are investing heavily in A.I. research and development, employing large teams of experts and contributing significantly to the advancement of A.I..

7. The open-source community plays a vital role in A.I. development, sharing code, datasets, and tools, fostering collaboration and innovation.

In essence, A.I. is being programmed by a diverse community of researchers, engineers, scientists, and domain experts, all working together to push the boundaries of what’s possible with this transformative technology.

Who’s really programming A.I.?

While A.I. development is concentrated in leading tech nations, the creation of A.I. content involves a global network of subcontractors and workers. This is especially true for tasks that require human input, such as data annotation, content moderation, and even generating training data.

Here are some of the nations frequently subcontracted to provide A.I. content:

India has a large pool of skilled workers and a thriving I.T. industry, making it a prime destination for outsourcing data annotation and other A.I.-related tasks.
The Philippines has a strong English-speaking workforce and a growing Business Process Outsourcing sector, making it attractive for content moderation and data labeling jobs.
Kenya has emerged as a hub for A.I. data annotation, with companies like Sama and Scale A.I. employing thousands of workers to label images, text, and other data.
Venezuela has a highly educated population and a struggling economy, making it a source of low-cost labor for A.I. related tasks.
Other countries in Africa, Southeast Asia, and Latin America are increasingly involved in providing A.I. content services due to lower labor costs and growing digital infrastructure.

Types of A.I. content tasks often subcontracted:

Data annotation: Labeling images, videos, and text data to train A.I. models.
Content moderation: Reviewing and filtering online content to ensure it meets platform guidelines.
Data collection: Gathering and cleaning data from various sources to be used for A.I. training.
Transcription and translation: Converting audio and text data into different languages. Maybe this is why image generators still cannot spell, even when given appropriate input.
Sentiment analysis: Analyzing text data to determine the sentiment or emotion expressed.

Business considerations:

Colloquial English skills may be lacking, leading to questionable data where colloquial responses are required.
Handling sensitive data requires robust privacy and security measures, which may not be enforced thoroughly in nations not friendly to the nations developing the technology. There are opportunities for data sifting and stealing.
Diverse datasets and hiring practices are crucial to avoid bias in A.I. systems. There are opportunities for data manipulation and unwanted easter eggs.

There’s a growing demand for transparency in A.I. supply chains, with companies being asked to disclose where and how their A.I. content is created. This can help ensure ethical practices and fair treatment of workers.

Today’s A.I. isn’t yet “intelligent.”

While A.I. has made incredible progress, here’s why current A.I. might not be considered truly intelligent:

Most A.I. systems excel at specific tasks but struggle to generalize knowledge to new situations or domains. They lack the flexibility and adaptability of human intelligence.
A.I. often processes information without true understanding. It can identify patterns and make predictions, but it doesn’t grasp the meaning behind the data in the way humans do.
Current A.I. lacks consciousness, self-awareness, and sentience. It doesn’t have emotions, desires, or a sense of self, resulting in responses that are based solely in pre-programmed datasets that may look clever and engaging, but aren’t based in anything that’s remotely intelligent.
A.I. relies heavily on vast amounts of data to learn. But it still has not learned to reason or make inferences in the same way humans can, especially in situations with limited data.

However, some argue current A.I. shows signs of intelligence:

A.I. systems can learn from feedback and adjust their performance over time. They can adapt to new information and adjust strategies.
A.I. can solve complex problems, often more efficiently than humans, in areas like logistics, finance, and healthcare. But the solutions are solely based on datasets, calculations, and programmer bias rather than considering ethics or the sanctity of life.
A.I. is being used to generate creative content, including music, art, and literature, pushing the boundaries of what machines can do. Most of the original creations are based on programmer bias, and are still super weird.

As A.I. continues to evolve, the line between artificial and human intelligence may become increasingly blurred. But for now, it’s important to recognize the limitations of current A.I. systems.

Levels of A.I.

There are several ways to categorize the levels of A.I., each with its own focus. Here are some of the most common frameworks:

1. Based on Capability:

Narrow or Weak A.I. (A.N.I.): This is the most common type of A.I. today. It’s designed to perform specific tasks, like writing articles, analyzing data, playing games, recommending products, or recognizing faces. A.N.I. excels in its domain but lacks general intelligence.
General or Strong A.I. (A.G.I.): This still hypothetical level of A.I. would possess human-like intelligence, with the ability to learn, understand, and perform any intellectual task that a human can. A.G.I. remains a long-term goal.
Super A.I. (ASI): This theoretical level surpasses human intelligence in all aspects. ASI could potentially solve complex problems beyond human comprehension, but it also raises concerns about control and ethical implications.

2. Based on Functionality:

Reactive Machines: These basic A.I. systems react to current situations without memory or the ability to learn from past experiences. Example: Deep Blue, the chess-playing A.I..
Limited Memory: These A.I. systems can store past data and use it to make better decisions. Examples: Self-driving cars, chatbots.
Theory of Mind: This future level of A.I. would understand human emotions and intentions, allowing for more natural interaction and collaboration.
Self-Aware A.I.: This hypothetical A.I. would possess consciousness and self-awareness, a level of intelligence that’s still largely in the realm of science fiction.

3. Based on Learning:

Supervised Learning: The A.I. learns from labeled data, where input and desired output are provided. Example: Image classification.
Unsupervised Learning: The A.I. identifies patterns and structures in unlabeled data. Example: Customer segmentation.
Reinforcement Learning: The A.I. learns through trial and error, receiving rewards for correct actions. Example: Game playing.

It’s important to note:

These levels are not always distinct, and there can be overlap between them.
A.I. is constantly evolving, and new levels and classifications may emerge as the technology advances.

ChatGPT vs Google Gemini

ChatGPT, and large language models like it, fall squarely into the category of Narrow or Weak A.I. (A.N.I.).

Here’s why:

Specialized Task: ChatGPT is designed for a specific task: generating human-like text based on the input it receives. It excels at this, but it doesn’t have general intelligence or the ability to perform a wide range of tasks like a human.
Lack of General Intelligence: While it can hold conversations, translate languages, and write different kinds of creative content, it doesn’t truly understand the meaning behind the words it generates. It lacks consciousness, self-awareness, and the ability to learn and adapt in the same way humans do.
Limited to Data: ChatGPT’s knowledge is limited to the massive dataset it was trained on. It can’t access real-time information or learn from new experiences in the way that humans can.

However, it’s important to acknowledge that ChatGPT pushes the boundaries of A.N.I.. It exhibits impressive capabilities in natural language processing and generation, making it seem more intelligent than traditional A.N.I. systems.

Here’s how it compares to other levels:

Not Reactive: ChatGPT is more advanced than purely reactive machines because it has some memory and can learn from the vast amount of text data it was trained on.
Not A.G.I. or ASI: It’s far from achieving general intelligence or superintelligence. It doesn’t have consciousness, emotions, or the ability to understand the world in the way humans do.

ChatGPT is a powerful example of Narrow A.I., but it’s important to remember its limitations. It’s a tool that can be used for various tasks, but it’s not a replacement for human intelligence or creativity. As A.I. technology continues to evolve, we may see systems that approach A.G.I., but for now, ChatGPT remains a sophisticated A.N.I. with a specific focus on language processing.

It’s tricky to definitively place Google’s Gemini on a specific level of A.I., as it has characteristics that span across different categories. But let’s break it down based on what we know:

Considering Capabilities:

Narrow A.I. (A.N.I.) with hints of A.G.I.: Gemini undoubtedly excels in a wide range of tasks, from language understanding and generation to code generation and image creation. This puts it firmly in the A.N.I. category. However, its multimodal capabilities and ability to learn and reason across different domains suggest it’s pushing the boundaries towards A.G.I., although it’s not quite there yet.

Considering Functionality:

Limited Memory with potential for Theory of Mind: Gemini clearly has limited memory, as it can access and process information from its vast training data. With its advanced language understanding, it can generate responses that consider different perspectives and emotional nuances. However, it’s not yet capable of truly understanding human emotions and intentions.

Considering Learning:

Combination of learning approaches: Gemini uses a combination of supervised, unsupervised, and reinforcement learning techniques. Its training involves massive datasets and complex algorithms that allow it to learn patterns, relationships, and generate novel outputs.

Key Takeaways:

Gemini is a highly advanced narrow A.I., demonstrating impressive capabilities across various domains.
Continual Evolution: As Google continues to develop Gemini, we can expect to see even more sophisticated capabilities emerge, potentially blurring the lines between A.N.I. and A.G.I..

It’s important to remember that A.I. is a rapidly evolving field. Gemini represents a significant step forward, but it’s still an early stage in the journey towards truly intelligent machines.

In 2025, what is A.I. useful for?

A.I. is rapidly changing the world around us, proving useful in a multitude of ways. Here are some of the key areas where A.I. excels:

1. Automation:

A.I. can automate mundane and repetitive tasks, freeing up human workers for more creative and strategic endeavors. Think of things like data entry, assembly line work, and customer service inquiries.
A.I. is crucial for self-driving cars, traffic optimization, and autonomous drones.
A.I. is used for medical diagnosis, drug discovery, personalized medicine, and robotic surgery.
A.I. powers fraud detection, risk assessment, algorithmic trading, and customer service.

2. Data Analysis and Insights:

A.I. can analyze massive datasets far beyond human capability, uncovering hidden patterns, trends, and insights. This is crucial in fields like finance, law, marketing, and scientific research.
A.I.-powered analytics tools help businesses and organizations make data-driven decisions, leading to better outcomes.

3. Personalization:

A.I. enables personalized experiences in areas like streaming audio and video
A.I. enables tailored learning plans for students learning, automated grading, and chatbots for student support.
A.I.-powered chatbots and virtual assistants provide instant and personalized customer support, enhancing productivity and lowering head counts.

4. Problem Solving:

A.I. can tackle complex problems in areas like healthcare, logistics, and environmental science. For example, A.I. can help diagnose diseases, optimize supply chains, and predict natural disasters.
A.I. algorithms can explore innovative solutions and approaches that humans might not consider.

5. Creativity and Innovation:

A.I. is being used to generate creative content, including strange music, really weird art, and writing. Proofreading and paraphrasing are two of the most useful applications of today’s A.I..
A.I. can assist in the development of enhanced technologies by analyzing data, generating ideas, and optimizing designs.

This is just a glimpse of what A.I. can do. As the technology advances, we can expect even more innovative applications that will transform our lives in countless ways. Regardless of the current state of the actual intelligence of A.I., Dharmesh Shah, CTO of Hubspot and Agent dot A.I., along with Jensen Huang, predict the future will not be just about singular GPTs, but more about networks or systems of programmatic A.I. agents where agents discover and collaborate with other agents.

In the same way that current large language models (L.L.M.s) support tool calling – where a set of tools is made available to the L.L.M., which it can use to formulate a more useful response – agents will be provided access to a network of other agents. The agent can then invoke these other agents in order to handle tasks that the subcontracted agents specialize in. Instead of being a Jack of all trades, your agent is free to become the master of one.

We’re already seeing this in powerful dev frameworks, building these kinds of multi-agent systems like CrewA.I. and low-code platforms like agent.ai.

2025 is going to be the year we see the first billion-dollar vertical A.I. agent companies emerge. Similar to what happened with vertical SaaS companies in the last decade. Instead of trying to build one-size-fits-all software, companies focused on specific industries and dominated them. Shah believes the same thing is about to happen with A.I. agents. There will be more specialized, independent A.I. agents that handle tasks involving collecting and synthesizing information, separate agents that can generate digital outputs (documents, emails, reports, etc.), agents that know how to use common software tools and platforms, and finally agents that can string these tasks together into useful workflows.

Any industry that heavily relies on these kinds of tasks — collecting information, creating digital content, using software tools — is going to be transformed.

Shah also believes A.I. agents are going to start changing how we interact with technology by becoming our digital proxies — understanding our preferences and handling some basic tasks similar to the way we would. He cites the memory of long gone travel agents, those humans who understood your travel preferences and handled all the complex details of planning a trip for you. Somehow, society went backwards and moved to doing everything ourselves — juggling dozens of websites to book a single trip.

Shah predicts we’re going to see the return of the “travel agent” model, but this time powered by A.I. agents that understand your preferences and patterns, followed by agents that can handle complex, multi-step shopping tasks without supervision. So instead of spending hours comparing flight prices, hotel reviews, and trying to coordinate dates, your personal A.I. agent will handle all of that. Just tell it “Plan me a week-long vacation to Japan in March within this budget” and it takes care of everything.

This same model could apply to finding and evaluating software for your business (like choosing a new accounting system), managing your personal calendar and scheduling meetings for you, and working with other people’s agents to get things done.

As we await true A.G.I. to kick in somewhere around 2028-2029, these A.N.I. agents will improve as they work with you and learn your particulars. They’ll learn your preferences, understand your habits, and become more effective over time.

In the meantime, skip using A.I. to generate images with any words, and use A.I. to help you write and analyze things as we watch an amazing automated and intelligent future unfold.