Anticipating Challenges in the Application of Neural Networks

                The application of Neural Network models is jarring as it outpaces people in what was historically considered human centric activities of work, such as making discoveries in science or at play,  demonstrably, in Chess or Go. These breakthroughs have thrown society into a convulsed layer of shock as to cycle between celebrating and feeling alarmed at the marvel of the only technology invented so far that we are not able to predict will remain in humanity’s control. Currently we assuage the need for concern that Artificial Intelligence cannot approximate human creativity or humanity as such those jobs may not be taken over by AI anytime soon.

                However, if AI is able to approximate intuition as displayed by Google DeepMind’s AlphaGo in its mastery of beating the best human player in Go (ColdFusion, 2016), the most complex human boardgames on Earth, how can we say with certainty it will not approximate creativity or sentiments or any of the technical or usage limitations we see at this time? Neural Networks theory has been around since the 1940s. The Perceptron is the first Neural Network built back in 1958 by Frank Rosenblatt (IBM, 2025). However, we did not see the amazing applications  possible for Neural Networks until the Internet brought access to massive data sources needed for training which also covered with the development of powerful microprocessors such as nVidia’s line of Graphics Processing Units to completing processing required to aide in scientific discoveries.

                People are admonished all the time to focus for success. We are also encouraged to work on our strengths as the payoff may prove more efficient than spending time and resources trying to develop a skill, we are not good at. In the same token, society sometimes celebrates jack-of-all trades as a budding entrepreneur may display. Looking at the confusion value, we put on specialization and general skills, what are AI developers to do in terms of focus and priority for creating general-purpose AI? It is inherent that despite the remarkable results and setbacks in current models, we will achieve general purpose AI because the neural network is based on the brain. As we get to understand the brain better, we will develop better models and improve neural network architecture to mimic general human intelligence. AI was in the valley for a while until the availability of large datasets and Graphics processing Units for computing power. Similarly, there is a discovery or development waiting in the wings to accelerate the development of Artificial General Intelligence models. 

                The quality of data impacts on the quality of the Neural Network Machine Learning model. Models that are trained in biased or incomplete corpuses of data will not be as effective for application. As such, we will have limited use of models for which we do not sufficient training data. Prepping data can lead to skewing if we leave information out  or fail to normalize units (AltextSoft, 2021). As such, we will have limited application of Neural Network for models that do not have sufficient training data such as medication for a rare disease. One way that Artificial Neural Networks may be able to get around the shortcomings of inadequate data (Stryker, 2025) is to generate synthetic data to learn or compete against itself to master a task. If a neural network model excludes persons based on sex, education, location, or other factors the model may penalize the weights that select these persons for a task.

                As such, ethics boards that create framework for neural network research, development and application are a critical part in the development of this field. From the Explore Activities it’s interesting that Google Alpha Go team believe their model has semblance of artificial intelligence while Professor Dutta feels that Neural Networks are challenged by reuse and knowledge transfer application to other domains. The field is going by leaps and bounds and weaknesses cited such as, transparency into how Neural Network make decisions, will be eventually get resolved by researchers or AI Models themselves.

AltexSoft. (2021, August 31). How is data prepared for machine learning?. YouTube.https://www.youtube.com/watch?v=P8ERBy91Y90&t=7sLinks to an external site.

ColdFusion. (2016, May 2). Google’s Deep Mind Explained! – Self Learning A.I. YouTube.https://www.youtube.com/watch?v=TnUYcTuZJpMLinks to an external site.

IBM. (2025, June 4). What is a neural network?. IBM. https://www.ibm.com/think/topics/neural-networksLinks to an external site.

Stryker, C. (2025, May 2). What is training data?. IBM. https://www.ibm.com/think/topics/training-data