Jrcb4 The Impact of Artificial Intelligence on Learning final

there may be fundamental theoretical and practical limits in designing AI 
systems that can explain their behaviour and decisions, it is important to keep 
humans in the decision-making loop. 
85
The quote is from the Manifesto of the Communist Party by Marx & Engels, 1848. 
86
There have been very few attempts to analyse AI from the point of view of learning theories. The learning 
capabilities of convolutional neural networks have been compared with Vygotsky's model of conceptual 
development in Tuomi (2018). 
87
Pattern matching systems can be very fragile in their decision-making capabilities. It is possible, for 
example, to fool image recognition programs by modifying image pixels (e.g., Yuan et al. 2017; Kurakin, 
Goodfellow, and Bengio 2016). 
88
Luckin (2018). 

---

36 
As several recent reports have emphasized, ethical considerations become highly 
relevant when AI is applied in the society or in educational settings.
89
From a policy 
perspective, the ethics of AI is a generic challenge, but it has specific relevance 
for educational policies. 
From the regulatory point of view, ethical considerations provide the fundamental basis 
from which new regulations and laws are created and justified. From a developmental 
point of view, ethics and value judgements underpin fundamental concepts such as 
agency, responsibility, identity, freedoms, and human capabilities. In supervised AI 
learning models, the possible choice outcomes need to be provided to the system before 
it starts to learn. This means that the world becomes described in closed terms, based on 
predefined interests and categories. Furthermore, the categories are based on data that 
are collected in the past. Neural AI categorizes people in clusters where data from other 
people, considered similar by the system, is used to predict individual characteristics and 
behaviour. 
From political and ethical points of view, this is highly problematic. Human agency means 
that we can make choices about future acts, and thus become responsible for them. 
When AI systems predict our acts using historical data averaged over a large number of 
other persons, AI systems cannot understand people who make true choices or who 
break out from historical patterns of behaviour. AI can therefore also limit the