The present is where decisions happen

The past gives us information.

The future gives us possibilities.

But there is only one place where we can actually do anything:

the present.

We cannot change an event that has already happened. We cannot directly act on an event that has not happened yet.

We can only act now.

This makes the present a particularly important point in any intelligent system. It is where information about the past is interpreted, predictions about the future are considered, and decisions become actions.

A useful way to think about this process is:

OBSERVATION → INTERPRETATION → DECISION → ACTION

And action changes the world.

From observation to action

Imagine you are approaching a pedestrian crossing in a car.

You observe that someone is standing beside the road.

You interpret their movement and decide that they may be about to cross.

You decide to slow down.

You press the brake.

The sequence is:

OBSERVATION → INTERPRETATION → DECISION → ACTION

This happens so naturally that we rarely think about its individual stages.

But each stage is different.

Observation

What information is available?

Interpretation

What does that information mean?

Decision

Given what we know, what should we do?

Action

How do we change the world?

This basic structure appears everywhere.

A doctor observes symptoms, interprets evidence, chooses a treatment and administers it.

A company observes demand, interprets market conditions, decides how much to produce and changes production.

A thermostat measures temperature, compares it with a target, decides whether heating is required and switches the heating system on.

A robot observes its environment, estimates what is happening, selects an action and moves.

Artificial intelligence increasingly participates in each of these stages.

The present is not just a moment in time

It is tempting to think of the present as simply the tiny boundary separating the past from the future.

For decision-making systems, however, the present has another meaning.

It represents the current state of the world as we understand it.

Imagine an autonomous vehicle.

At a particular moment it might need to know:

where it is,
how fast it is travelling,
which direction it is moving,
where nearby vehicles are,
where pedestrians are,
what the traffic lights are showing,
what route it is following,
how much energy it has available.

Together, information like this describes the system's current situation.

We can call this its state.

The state gives us a starting point from which to reason about what might happen next.

Observation is not the same as state

There is an important distinction here.

We rarely observe the complete state of the world directly.

Instead, we receive observations.

A camera produces pixels.

A thermometer produces a temperature measurement.

A GPS receiver produces an estimated location.

A microphone produces an audio signal.

A electricity meter produces measurements of electrical quantities.

These are observations.

An intelligent system may then use those observations to construct an internal representation of what it believes is happening.

Consider a robot looking at a road.

The camera does not directly tell the robot:

A cyclist is travelling towards me at 18 km/h and is likely to cross my path in three seconds.

The camera provides images.

The system has to interpret those images.

It might identify an object as a cyclist, estimate the cyclist's position, estimate their velocity and infer their likely trajectory.

The observations therefore help the system estimate the state of its environment.

This distinction will become extremely important when we later examine state vectors, state estimation, robotics and autonomous systems.

Interpretation gives data meaning

Raw observations do not necessarily tell us what to do.

They have to be interpreted.

Suppose a temperature sensor reports:

38.7°C

What does that mean?

For the outside temperature in Dublin, it would be extremely unusual.

For the temperature of a human body, it could indicate a fever.

For the temperature inside an industrial process, it might be perfectly normal.

For a computer processor, it might be relatively cool.

The number is the same.

Its meaning depends on context.

Interpretation connects observations to a model of the thing being observed.

This is one of the roles increasingly performed by artificial intelligence.

A computer-vision model interprets pixels.

A speech-recognition model interprets sound.

A language model interprets sequences of tokens.

A forecasting model interprets historical patterns.

But interpretation still does not tell us what action should be taken.

For that, we need a decision.

Prediction helps us decide

Many decisions depend not only on what is happening now, but on what we believe will happen next.

Consider the pedestrian crossing again.

You do not brake because the pedestrian has already collided with the car.

You brake because you believe that continuing at the current speed could produce an undesirable future.

Your decision depends on a prediction.

The structure therefore becomes:

OBSERVATION → INTERPRETATION → PREDICTION → DECISION → ACTION

We observe the current situation.

We interpret what those observations mean.

We consider possible future outcomes.

We choose an action.

This is one of the central roles of prediction.

Prediction gives decisions information about possible futures.

But prediction and decision are not the same thing.

A prediction tells us what might happen.

A decision determines what we should do about it.

We will return to this distinction repeatedly throughout the course.

Decisions require objectives

Suppose a navigation system predicts that three routes will take:

Route	Predicted journey time
A	24 minutes
B	27 minutes
C	31 minutes

If the only objective is to minimise journey time, the decision appears obvious.

Choose Route A.

But what if Route A has a toll?

What if Route B uses less fuel?

What if Route C avoids a school during morning drop-off?

What if thousands of other navigation systems are simultaneously directing drivers towards Route A?

The prediction alone cannot determine the best decision.

We need to know what we are trying to achieve.

That means defining an objective.

Possible objectives might include:

minimise time,
minimise cost,
minimise energy consumption,
maximise safety,
maximise profit,
minimise emissions,
maximise reliability,
improve fairness.

Real systems frequently have several objectives at the same time.

And those objectives can conflict.

Decisions also face constraints

We cannot simply choose any imaginable action.

The world imposes constraints.

A delivery vehicle cannot instantly move from one city to another.

A battery cannot provide more energy than it contains.

A hospital cannot allocate a bed that does not exist.

A power line cannot safely carry unlimited electricity.

A person cannot spend the same euro twice.

A robot cannot move through a wall.

So decision-making involves more than asking:

What would we like to happen?

It also requires asking:

What actions are actually possible?

This distinction between objectives and constraints will become central when we study optimisation and the allocation of shared resources.

Decisions alter the state of the world

This is where things become particularly interesting.

A decision is not merely a passive response to the world.

Once acted upon, it changes the world.

Suppose the current state is:

STATE NOW

We make a decision and take an action:

STATE NOW → DECISION → ACTION

The action changes something.

We then arrive at:

NEW STATE

So we can write:

CURRENT STATE → ACTION → NEXT STATE

The next state becomes the starting point for the next decision.

This creates a continuous process:

STATE → OBSERVE → DECIDE → ACT → NEW STATE → OBSERVE → DECIDE → ACT → ...

This is the basic structure of many dynamic systems.

The future depends on what we do

This has an important consequence for prediction.

The future is not necessarily something that simply happens to us.

Our actions can change it.

Suppose a model predicts that a machine is likely to fail tomorrow.

A maintenance team replaces the failing component today.

Tomorrow arrives.

The machine does not fail.

Was the prediction wrong?

Not necessarily.

The prediction may have caused the action that prevented the predicted future from occurring.

This creates an interesting relationship between prediction and intervention.

A sufficiently useful prediction can help us avoid the future it predicts.

Predictions can change behaviour

The effect becomes even more interesting when predictions are communicated to people or other machines.

Imagine a navigation system predicts congestion on Road A.

It tells thousands of drivers to use Road B instead.

Drivers respond.

Road A becomes quieter.

Road B becomes congested.

The prediction has influenced the system being predicted.

Something similar can happen in:

financial markets,
electricity systems,
transport networks,
online recommendation systems,
supply chains,
social networks.

A prediction is no longer simply observing the system from outside.

It can become part of the system itself.

The feedback loop

This gives us a more complete picture of intelligent decision-making:

OBSERVATION → INTERPRETATION → PREDICTION → DECISION → ACTION

But the action changes the world.

So we need another arrow:

ACTION → NEW STATE → NEW OBSERVATION

The complete process becomes a loop:

OBSERVATION → INTERPRETATION → PREDICTION → DECISION → ACTION → NEW STATE → OBSERVATION → ...

This is called feedback.

Feedback is one of the most important ideas in engineering, biology, economics, robotics and intelligent systems.

An intelligent system does not necessarily make one prediction and one decision.

It observes what happens after its action and then decides again.

A thermostat is an intelligent loop

Consider one of the simplest examples: a thermostat.

Suppose we want a room to remain at 20°C.

The thermostat observes:

18°C

It interprets this relative to the desired temperature:

too cold

It decides:

heating required

It acts:

turn heating on

The room begins to warm.

The state of the world changes.

The thermostat measures again:

19°C

Then:

20°C

Now the decision changes:

turn heating off

The system repeatedly observes and acts.

It does not need to perfectly predict the temperature of the room for the next 24 hours.

It needs to continuously respond to the state of the system.

This simple feedback structure will later help us understand much more sophisticated systems involving control, optimisation, artificial intelligence and robotics.

AI increasingly sits inside the loop

Traditional machine-learning systems were often used primarily for prediction.

Give the model some data.

Receive a prediction.

A human then decides what to do with it.

For example:

DATA → AI PREDICTION → HUMAN DECISION → ACTION

But increasingly, AI systems participate in more of the loop.

An autonomous system might:

observe its environment,
interpret those observations,
estimate the current state,
predict possible future states,
evaluate possible actions,
choose an action,
act,
observe the result,
repeat.

The human decision-maker may no longer sit between every prediction and every action.

This is a profound transition.

It moves AI from being merely a tool for prediction towards becoming part of a system of agency.

Faster loops create different systems

The speed of this loop matters.

A government might:

collect data → analyse it → make policy → observe outcomes

over months or years.

A business might adjust prices every day.

An electricity system might respond to changing conditions every few seconds.

A robot might update its understanding of the world many times per second.

As the loop becomes faster, there is less time for human intervention between observation and action.

That creates opportunities.

Machines can react far faster than humans.

But it also creates risks.

A poorly designed automated system can make incorrect decisions repeatedly and at enormous speed.

Automation therefore changes not only who makes decisions, but also the timescale at which decisions occur.

Decisions have consequences

This brings us to a question that will run throughout the course:

Who decides what the machine should decide?

An AI system does not magically discover what society wants.

Somebody has to define:

what information it receives,
what outcomes it predicts,
what objective it pursues,
what constraints it must respect,
which errors matter most,
who receives priority,
when humans can intervene.

These choices are part of the design of the system.

And once an intelligent system begins making decisions that affect people, those design choices become social choices as well as technical ones.

The present connects past and future

We can now extend the framework from the previous lesson.

The past provides observations:

PAST → DATA

Models help us interpret those observations:

DATA → MODEL

Models allow us to reason about what might happen next:

MODEL → PREDICTION

Predictions inform decisions:

PREDICTION → DECISION

Decisions become actions:

DECISION → ACTION

Actions change the world:

ACTION → NEW STATE

And once that new state is observed, it becomes new data.

So the full loop becomes:

PAST → DATA → MODEL → PREDICTION → DECISION → ACTION → NEW STATE → DATA → ...

The present is the point at which this loop can be influenced.

It is where information becomes action.

The central idea

The past cannot be changed.

The future cannot yet be observed.

But the present gives us something neither of them can:

the ability to act.

We observe the world as it has become.

We interpret those observations.

We reason about what might happen next.

We make decisions.

And through our actions, we change the state from which the next future emerges.

The past gives us evidence. The present gives us agency. The future gives us possibilities.

Artificial intelligence makes this process increasingly powerful because machines can observe more information, recognise more complex patterns, generate predictions more quickly and increasingly participate directly in decisions and actions.

But no matter how sophisticated the technology becomes, the basic structure remains:

OBSERVE → INTERPRET → PREDICT → DECIDE → ACT → OBSERVE AGAIN

In the next lesson, we turn our attention to the other side of the present:

the future — something we cannot observe, can never know with complete certainty, but must nevertheless make decisions about.