Van het Beschermen van Walvissen tot Meer Gemak, Wat Kan AI Niet Doen?

Kunstmatige intelligentie wordt steeds gewoner in het dagelijks leven. De komst van LLM’s (grote taalmodellen), zoals ChatGPT, bracht AI binnen ieders bereik. De technologie was echter al hard op weg om levens te veranderen, lang voordat LLM’s op de markt verschenen. Hieronder volgt een blik op wat AI momenteel nog niet kan en waar de technologie in de toekomst kan uitkomen.

AI Evolutie

The AI sector has undergone numerous iterations based on the technology’s capabilities and accessibility. The concept of human-like machines dates back to ancient times. The Renaissance revived these concepts, and famous inventors like Leonardo da Vinci demonstrated the concept with automatons in 1495.

Kunstmatige intelligentie zoals we die vandaag zien, kan worden herleid tot Alan Turing’s Universele Turingmachine in 1935. Slechts 15 jaar later ontwikkelde hij de Turingtest. De Turingtest gebruikt menselijke conversatie om te bepalen of een AI-systeem mensachtige capaciteiten en meer heeft bereikt. Deze test wordt nog steeds vandaag gebruikt als een maatstaf voor AI-interacties.

The 1990s computational breakthroughs helped to push AI systems further. Machine learning and other AI systems began to emerge during this time, leading to breakthroughs in the early 2000s. For example, IBM demonstrated Watson’s AI skills by beating out humans in a game of jeopardy. At this same time, Google started releasing data on its neural network systems.

AI Vandaag

This decade has seen some of the greatest leaps in AI technology so far. The introduction of self-learning AI technologies like torque clustering algorithms has made the sector a point of discussion.

These technological gains have been met with moral questions and environmental concerns as well. All of these factors have led to more scrutiny of the future of AI and its influence on daily lives. So what can’t AI do? These projects highlight just some of its current capabilities and limitations.

Mensen Helpen Betere Steden te Bouwen

AI-systemen kunnen stadsplanners helpen om beter functionerende en efficiëntere gemeenschappen te creëren. Ze kunnen het verkeer en andere belangrijke gegevens volgen die van grote waarde zijn bij het bepalen van locaties voor nieuwe voorzieningen, snelwegen en andere behoeften van de gemeenschap.

Infrastructuurinzichten

AI-systemen worden nu al ingezet om buurten veiliger en comfortabeler te maken door inzichten uit gegevens te benutten. Een studie¹ die in het Journal of Smart Cities and Society is gepubliceerd, onderzoekt de veiligheid en stabiliteit van elektriciteitsnetten diepgaand.

Specifiek onderzoekt het de energiedistributie, woningvereisten, welk type energiebron elk huis gebruikt, en of dat de locatie een hoger risico op stroomuitval geeft. De studie legt uit dat naarmate meer huizen overstappen op volledig elektrische opties, sommige risico’s over het hoofd kunnen worden gezien.

In particular, the report highlights how solar panels are a great source of energy during the summer, but how they can leave entire communities without power during winter storms and other environmental conditions. Sadly this scenario has become more common as the push for all electrified homes continues.

Een AI-model Bouwen om Risico op Stroomuitval te Rangschikken

Stevens researchers analyzed data from the Department of Energy (DOE) building stock, including the energy patterns of 129,000 single-family homes. These houses were located in eight different states, allowing the team to test the system in various environments.

The ML models successfully determined individual house energy system footprints. It then processed this data, cross-referenced it with its data models, and determined what type of energy sources the house relied on.

Furthermore, the system took this data and used it to decide the blackout risks to the house and community. Now, the team seeks to expand its testing to more communities to help planners build safer and more resilient electrical grids.

Red de Aarde

Environmental conditions, pollution, human expansion, and other factors have left many wildlife regions in tatters. Artificial intelligence has been one solution that environmentalists have turned to to help determine wildlife patterns.

These systems can help conservationists notice sudden changes in species, which is a valuable tool that can assist in protecting rare and endangered animals from extinction. Already, AI systems can track migration patterns, health, poacher activity, and other vital data.

Walvisbescherming

A team of innovative researchers from Rutgers University-New Brunswick introduced an AI model designed to help international shipping vessels avoid endangered species populations recently.

De hoofdauteurs van de studie², Ahmed Aziz Ezzat en Josh Kohut, maakten het model voor de migratie en het volgen van walvissen met uiteenlopende gegevens om de zeldzame Noord-Atlantische noordkaper te beschermen. Helaas blijkt uit gegevens van de Amerikaanse nationale dienst voor oceanen en atmosfeer dat er minder dan 400 van deze indrukwekkende dieren in het wild leven. Van de resterende populatie zijn slechts 70 vrouwtjes vruchtbaar, wat natuurbeschermers extra zorgen baart.

Bron – Nature Communications

Specifically, the team integrated the Rutgers University Center for Ocean Observing Leadership data from as far back as the early 1990s, satellite imagery from the University of Delaware, and underwater glider info into the model. Notably, gliders are underwater craft that have an array of sensors. They traverse the ocean floor helping to map uncharted regions, track environmental changes, and search for valuable resources,

The AI predicted the location and time of whale populations based on the whale’s preferences, past locations, environmental conditions, and time of year. It enabled researchers to connect these dots and find the patterns that increased whale sighting possibilities. In the future, this system will accurately show ships where whale populations are, enabling them to rechart a course avoiding these natural habitats.

Kunstmatige Intelligentie van Dingen (AIoT).

The melding of Artificial Intelligence and Internet of Things technologies created the now budding AIoT industry. This sector combines the reach and data gathering capabilities of IoT systems with pattern recognition, processing, and convenience AI systems provide.

AIoT Systems enable logistics companies to get their products across the globe faster, reduce counterfeiting, and provide a host of other high-level features that would be impossible. AIoT improves on the IoT concept in a couple of key ways including efficiency.

Traditional IoT systems operate as sensors. They gather and send data via the internet to another location that then processes that data.  AIoT eliminates the need to send the data over the internet. Instead, these systems can process the information in-house, reducing time, costs, and bandwidth requirements.

Slimme Economie

AIoT will serve a crucial role in tomorrow’s smart economy. Today’s smart homes rely on various human activity recognition protocols to try and provide convenience to their operators. For example, imagine telling Alexa that you want to work out, and initiating a sequence of tasks designed to make the process more convenient.

These tasks could include adjusting the lighting, playing music, closing the blinds, and even starting up the after-workout smoothie maker. All of this is possible now using preset macros. However, the use of advanced human activity recognition AI systems will automate this process.

MSF-Net

Incheon National University engineers recently raised eyebrows after introducing³ their MSF-Net (multiple spectrogram fusion network) Wifi human activity tracking system. The system can leverage channel state information (CSI) to determine subtle human activity.

This technology will make smart homes more convenient. Soon AI will be able to determine if you are cooking, resting, watching TV, or about to leave for work and adjust the environment accordingly. This technology could see integration into multiple sectors including manufacturing, security, and healthcare.

Moreel Advies

Hoe geavanceerd AI-technologie ook wordt, er lijkt altijd enig wantrouwen tegenover deze systemen te blijven bestaan. Een recente studie⁴ in het tijdschrift Cognition laat zien dat mensen AI-systemen niet vertrouwen als het om moreel advies gaat.

The School of Psychology studies why there is a distrust of AI systems, how it has slowed adoption, and what it would take to overcome these roadblocks to adoption. The researchers looked at the rise of artificial moral advisors (AMAs) and how they could one day become common.

Artificial moral advisors are purpose-built algorithms that leverage pre-programmed ethical theories, principles, and guidelines to provide users with answers to difficult moral dilemmas.

Notably, the study found that people were wary of AI advice on moral issues, even when it was the same advice given by a human advisor. It also demonstrated that many people have misconceptions about AI and concerns regarding its ability to go rogue.

In the future, you may have an AMA by your side to help you make the right choice. At the very least, it will document and probably report any “wrong” choices you make. As such, it’s easy to see why humanity may be ready to give AI missile launch codes, but not let it run the sermon.

Wat kan AI niet doen? Er is nog veel

The AI revolution is in full swing and major technological breakthroughs continue to drive AI system capabilities further. Despite the constant echo that AI is coming for your job, most people still have plenty of time before their robot overlords take control. However, the human touch is still a way out.

Many factors like distrust and misunderstandings will inevitably lead to skepticism regarding AI’s true capabilities and purpose. For now, this tech has infinite potential but remains limited in its ability to connect to humans. As LLMs and robotics improve, the line between humanity and AI may meld further, opening the door for new realities.

Learn about Other Cool AI Projects Now

Studieverwijzing:

^{1. Majowicz, A., Popli, C., & Odonkor, P. (2025). Quantifying household vulnerability to power outages: Assessing risks of rapid electrification in smart cities. Journal of Smart Cities and Society, 0(0). https://doi.org/10.1177/27723577241306340}

^{2. Ji, J., Ramasamy, J., Nazzaro, L., Kohut, J., & Ezzat, A. A. (2024). Machine learning for modeling North Atlantic right whale presence to support offshore wind energy development in the U.S. Mid-Atlantic. Scientific Reports, 14, 29147. https://doi.org/10.1038/s41598-024-80084-z}

^{3. Chen, J., Xu, X., Wang, T., Jeon, G., & Camacho, D. (2024). An AIoT framework with multimodal frequency fusion for WiFi-based coarse and fine activity recognition. IEEE Internet of Things Journal, 11(24), 39020-39029. https://doi.org/10.1109/JIOT.2024.3400773}

^{4. Myers, S., & Everett, J. A. C. (2025). People expect artificial moral advisors to be more utilitarian and distrust utilitarian moral advisors. Cognition, 256, 106028. https://doi.org/10.1016/j.cognition.2024.106028}