AGI

This year will mark 40 years since The Terminator movie hit the cinema screens. Also another Hollywood icon, The Matrix, will mark its 25-year anniversary. Both films gave birth to decades-long franchises that delivered tremendous commercial success. This reflects not just the talent of the movies’ creators, but also the worldwide public’s keen interest in the topic on which both films are based: The development of AI and its potential risks to humanity.

Well, in reality, AI has not reached the potentially dangerous phase of self-consciousness and is not anywhere close to it. It seems like we are in the very beginning of an AGI era, known as Artificial General Intelligence, opened by the launch of OpenAI’s GPT-4. Also, along with the not-so-obvious risks, AI reveals tremendous potential to improve our life and provide fantastic opportunities, especially for the startup community.

Humans vs. Machines: Has the Competition Started?

The Three Stages of Artificial Intelligence

ANI: Artificial Narrow Intelligence –  Machine Learning

AGI: Artificial General Intelligence – Machine Intelligence

ASI: Artificial Super Intelligence – Machine Consciousness

 

The term ‘AI’ itself is not new; it was coined in 1956, almost 70 years ago. Previously, we were primarily confined to ANI, which is closely related to machine learning. However, in March 2023, OpenAI introduced their GPT-4 model. I think that this event can be regarded as an initial step into the era of AGI, or artificial general intelligence, because it starts bridging the gap between human common sense and AI capabilities.

According to Sam Altman, the CEO of OpenAI, AGI is defined as an AI system equivalent to a median human, capable of effectively working alongside humans as a co-worker. However, achieving full-scale AGI involves creating AI systems that possess human-like intelligence and the ability to perform a wide range of tasks with adaptability and understanding of context. While there have been significant advancements in AI, particularly in areas like natural language processing and computer vision, AGI remains a long-term and challenging goal. Moving towards AGI is a complex endeavor that involves not only technical advancements but also ethical and safety considerations, which must be carefully addressed as we move forward.

Human Intelligence vs AI

Human Intelligence AI
Number of Neurons 86 billion Artificial neural network with 175 billion parameters (GPT-3)
Processing Speed 1 to 100 meters per second Trillions of floating-point operations per second,

e.g. GPT-4 Turbo can process a 300-page book within a second

Energy Efficiency 20 watts of power 20 million watts (GPT-3)
Complexity and Versatility Winner for the time being Operates in different modalities

 

Note: OpenAI has not officially disclosed the number of ANNs in GPT-4 yet. However, they’ve recently released an upgraded version called GPT-4 Turbo, which boasts enhanced capabilities and world knowledge up to April 2023. Notably, one of its impressive features is the support of a 128k context window, enabling it to accommodate the equivalent of over 300 pages of text in a single prompt.

Opportunities for New Players

Being realistic, I suggest that we do not attempt to look too much into the 23rd century, the era supposedly portrayed in The Matrix, or even the 2029 pictured in The Terminator. AI has become a very fast-developing industry, so long-term forecasts may be misleading. Latest research suggests that the AI industry has reached its plateau for now – probably preparing another boost soon. Training large AI models is becoming more and more expensive, while improvements are getting more and more marginal.

Amidst the current tech race and flatter technical performance, leaders in cutting-edge AI research change frequently. Researchers from EpochAI note that the overall leadership of tech majors like Google and Meta is obvious, new AI labs emerge and quickly take their place among the top innovators, like OpenAI. Therefore, future breakthroughs may come from players that are not commonly known or even do not yet exist at the moment.

AGI Tomorrow and Every Day

AI has a dual nature, much like a coin with two sides. While we are witnessing remarkable advancements in AI technology, particularly in the era of AGI, it’s important to acknowledge that along with the benefits, there come potential security concerns. Consider this scenario: What if an AI were to launch an attack on a critical sector like the nuclear industry or other vital infrastructure? Such an event wouldn’t merely result in the loss of a few lives; it could lead to the devastation of an entire city or even an entire country. However, investments in AI security and the associated concerns have been relatively scarce. The most recent EU regulation represents just the initial steps taken in this direction this year.

I believe that, in line with human nature, we can explore opportunities that not only promise substantial returns but also ensure the safety of humanity. Profitable and protecting humans – a win-win game, right? So, let’s dive into these opportunities in the AGI era, with a particular focus on No.8, which involves harnessing AI to safeguard us from potential AI threats. Let me provide you with some insight into the exciting opportunities for each subfield of AI:

  • Machine Learning is most commonly utilized in recommendation systems such as Netflix. I have to admit there is still a long way to go in terms of accuracy. With time, more such systems will emerge. Their predictive power will increase and their applications will be much broader than entertainment.

So, these are opportunities for us – not limited to machine learning AI technology, involving human-AI interaction such as ChatGPT and Whisper V3 as a brilliant image recognition model. We could invent a super intelligence recommendation system, which has the potential to revolutionize various fields, including academia. Here are some advanced examples of what such a system could achieve:

  1. Personalized Curriculum Design: The system could analyze a student’s educational history, learning style, and career goals. It would then recommend a customized curriculum, including the most relevant courses, textbooks, and research papers, ensuring students receive the best education tailored to their needs.
  1. Real-time Literature Analysis: This system could continuously monitor and analyze the latest academic publications, identifying emerging trends and breakthroughs in various fields. It would provide researchers with up-to-date information and suggest which papers to prioritize for reading.
  1. Language and Cultural Adaptation: For international students, the system could help bridge language and cultural gaps by recommending language courses, cultural sensitivity training, and resources for adapting to a new academic environment.
  • NLP is now known for powering virtual assistants, such as Alexa, Google Assistant and Siri.

Thinking broadly, it can also be employed in conjunction with other AI models, such as IoT smart home systems, to control physical objects with voice commands. Here are a few examples:

  1. Archaeological Exploration: Picture an archaeological expedition where NLP-driven AI assists researchers. Archaeologists can speak to AI-powered drones to identify potential excavation sites, and the AI can analyze ancient texts and hieroglyphics in real-time to uncover historical secrets buried beneath the earth.
  1. Space Exploration: Envision astronauts on a distant planet conversing with a language model that understands and translates alien languages in real-time. This AI facilitates peaceful interactions with extraterrestrial beings, enabling interstellar diplomacy and collaboration on a cosmic scale.
  1. Art and Creativity: Consider an AI-driven art studio where artists collaborate with intelligent machines. Artists can describe their creative vision in natural language, and the AI translates these descriptions into stunning visual or musical compositions, pushing the boundaries of human-machine creativity.
  • Expert Systems, such as IBM Watson are increasingly widely used for medical diagnosis.

As soon as they are integrated into a robust DALL-3-like image recognition model, they can also be used in surgery. Furthermore, when combined with a 3D printer and robotics AI technology, they can perform tasks such as bone surgery and even more complex procedures like neurosurgery with zero errors while simultaneously scanning neurons. Such systems would conduct neurosurgery in the fastest and safest manner, ultimately saving patients and prolonging their functional lives. Here are some more advanced applications:

  1. Nano-Surgeons: Imagine nanobots equipped with AI and 3D printing capabilities. These tiny robotic surgeons could navigate the bloodstream to perform precise surgeries at the cellular or even molecular level, such as removing cancerous cells without invasive procedures.
  1. Surgical Artist: Combining AI’s artistic creativity with surgical precision, an AI-driven robotic surgeon could perform intricate surgeries with a touch of artistic flair, potentially incorporating patient-specific designs or patterns into surgical procedures.
  1. Underwater Surgery: For marine explorations, imagine AI-driven underwater robots equipped with surgical tools, capable of performing surgery on aquatic creatures or even repairing coral reefs to protect underwater ecosystems.
  • Image recognition and machine vision platforms are currently used for quality check at manufacturing lines.

However, we can add an AI prediction model to predict production quality and also reduce workforce injury. Maybe we can think of more uses for this technology:

  1. AI-Enhanced Workers: Workers in smart factories wear augmented reality (AR) glasses that provide them with real-time information and guidance. The glasses use AI and machine vision to recognize machinery, display relevant data, and offer step-by-step instructions for complex tasks, enhancing worker productivity and safety.
  1. Eco-Friendly Manufacturing: AI predicts the environmental impact of production processes and suggests eco-friendly alternatives. For instance, it might recommend using recycled materials, optimizing energy consumption, and reducing waste to support sustainable manufacturing.
  1. Smart Materials: Manufacturing lines incorporate AI-driven quality checks for materials in real-time. When subpar materials are detected, AI can suggest adjustments to the production process to compensate for material deficiencies, reducing waste and maintaining quality.
  • Speech AI technology is mostly known for its application in speech-to-text (Teams, Zoom) and text-to-speech solutions (Play.ht).

Further on, we can use machine learning and whisper V3 to make video files or streams with customized audio tracks. Also, it is possible to integrate a variety of speech models into one solution to simplify user experience. For example, this super powerful tool can be employed in the content creation and advertising industry. Wow, astonishing! But still, there can be more options:

  1. Emotion-Responsive Ads: Advertisements can adapt their tone and content based on the viewer’s emotional cues. AI analyses facial expressions and vocal intonations in real-time to create ads that resonate with viewers on a deep emotional level.
  1. Virtual Shopping Assistants: In virtual reality shopping experiences, AI-driven virtual shopping assistants guide users through stores, provide product information, and even have casual conversations about user preferences, creating a personalized shopping journey.
  1. AI-Generated Concerts: Musicians and artists can collaborate with AI to create virtual concerts where the AI generates music, lyrics, and visual effects in real-time. These immersive experiences can be customized for each viewer.
  • Planning solutions are mostly used for self-driving cars to navigate their routes.

If integrated with a new smart recommendation system (yet to be developed), they can let us instantly adjust our plans to our calendar. Check out some more ideas:

  1. Seamless Calendar Integration: Imagine a world where your planning solution is seamlessly integrated with your calendar app. As you go about your day, the system continually analyses your calendar events, traffic conditions, and other real-time data.
  1. Adaptive Travel Routes: When your planning solution detects a potential delay or change in your schedule, it proactively suggests alternate routes or modes of transportation. For instance, if a meeting is running late, the system could recommend switching from your car to a nearby subway line for a faster arrival.
  1. Proactive Health and Well-being: The smart recommendation system can factor in your health and well-being goals. If it notices you’ve been working long hours, it may recommend scheduling short breaks or adjusting your exercise routine to keep you on track with your fitness goals.
  • Robotics technology got famous thanks to Boston Dynamics’s robot dog.

If these platforms integrate with DALL-3, Whisper V3, and ChatGPT, they can be more effectively applied in scenarios like hospital guidance or more:

  1. Genomic Supercomputers: Advanced robotic systems, integrated with DALL-3’s deep understanding of genomics, can rapidly sequence and analyze entire genomes in real-time. This accelerates personalized medicine, enabling on-the-spot diagnosis and treatment plans tailored to a patient’s genetic makeup.
  1. Robotic Pharmacies: Automated pharmacy robots, backed by ChatGPT for patient consultations, can dispense medications, provide detailed medication instructions, and offer immediate answers to patients’ questions. These robotic pharmacies operate 24/7, ensuring access to medications at any hour.
  1. Rapid Epidemic Response: In the face of a pandemic or epidemic, swarms of medical robots equipped with real-time disease detection capabilities work in tandem with human healthcare workers. They can quickly diagnose, isolate, and treat infected individuals, while ChatGPT provides public health education and reassurance.
  • Leverage AI Against AI, anything that can potentially distinguish AI from human beings, like recognizing the text and images generated by AI, is crucial. This presents a wide range of untapped opportunities that not only push AI boundaries but also ensure human safety in the AGI era. Let’s explore some illustrative examples of how this can be achieved:
  1. AI-Powered Fact-Checkers: Advanced AI systems can be developed to fact-check and verify information, flagging AI-generated content that lacks credibility. This ensures that fake news, misinformation, and deep fakes are quickly identified and mitigated.
  1. AI Transparency Standards: Developing industry standards and regulations for AI-generated content disclosure and transparency can help users differentiate between human and AI-generated contributions, especially in creative works and journalism.
  1. AI in Education: Educational institutions can use AI to detect AI-generated essays or assignments submitted by students. This ensures academic integrity and encourages originality in learning.

In this field, there is a blue ocean in front of us and this technology is crucial, as it can protect human beings against AI products in the AGI era.

Conclusion

Actually, our fears of a ‘machine rebellion’ may well speak more about ourselves than about the true risks of AI. Humans are an aggressive, dominant biological species, these features being forged by millions of years of evolution; this ultimate race for life. Our greatest existential fear is to encounter a competitor that surpasses us in cognitive power, which has always been our key to survival and success.

There are no indications that an AI created by humans can inherit our aggressiveness, or a goal to fill any available space with itself and its siblings, typical to most living things. But, of course, there are also no indications that AI will be peaceful and humane, which is also part of a human mind’s nature. Most probably, it will be just different. Anyway, we cannot know the answer and it may actually take a long while until we do.

As history suggests, technological progress, invented by humans, cannot be stopped by humans. So despite all the odds and risks, we should live with it, find our place and new opportunities in the new and ever changing reality. I hope that you all can catch the ‘AI train’ to become the next FAANG firms in the AGI era, a global shift similar to the start of the Internet era twenty years ago.

Last but not least, we urgently need more people to join the field of AI Security technology. Let’s protect humanity and protect ourselves in the AGI time.