Blog 4: Shadows of the Mind - MDP in Mental Health

Scene 1: A Morning Without Light

Imagine waking up and feeling like the world has no color. You look at your phone, no message excites you. Breakfast tastes bland. Work feels pointless.

Every action that once brought joy - meeting friends, finishing a project, listening to music - now feels empty.

This is not laziness. This is depression. And in another corner of the mind, there's anxiety: Your brain imagines worst-case scenarios, even when nothing has gone wrong yet. A simple text - "we need to talk" - spirals into catastrophic outcomes.

Both conditions may seem emotional and mysterious, but beneath them lies a very mathematical truth: our brains' decision-making engine - the Markov Decision Process - gets distorted.

Scene 2: Depression as a Low-Reward MDP

In depression, the world looks like an environment with poor rewards.

  • States (S):
    • At home, at work, with friends.
  • Actions (A):
    • Go out, call a friend, start a task, stay in bed.
  • Rewards (R):
    • Normally: calling a friend → +10 happiness.
    • In depression: calling a friend → +0 (flat, no joy).

So the brain updates its Q-values:

$$Q(\text{call friend}) \approx 0$$ $$Q(\text{stay in bed}) \approx 0$$

Everything starts to look equally unrewarding. Over time, the brain "learns" that no action is worth taking. This is why people with depression often say: "What's the point?"

Scene 3: Anxiety as a Threat-Biased MDP

In anxiety, the problem is not low reward - it's overestimated risk.

  • States (S):
    • Boss calls you for a meeting.
  • Actions (A):
    • Enter calmly, prepare logically.
    • Overthink, avoid, panic.
  • Rewards & Transitions (P):
    • In reality: 80% chance it's routine, 20% chance it's criticism.
    • In anxiety: brain rewires probabilities → 90% chance it's negative.

Mathematically:

$$P(\text{criticism} | \text{meeting}) = 0.2 \quad \text{(real)}$$ $$P(\text{criticism} | \text{meeting}) = 0.9 \quad \text{(anxiety brain)}$$

So the expected value of attending the meeting feels negative. Avoidance seems rational - even though it's not.

Scene 4: The Bellman Equation in a Dark Mirror

Normally, the Bellman equation is a powerful guide:

$$V(s) = R(s,a) + \gamma \sum_{s'} P(s'|s,a) V(s')$$

But in depression:

  • \(R(s,a) \to 0\). Rewards collapse.

In anxiety:

  • \(P(s'|s,a)\) becomes skewed toward negative outcomes.

In both cases, the recursive evaluation of life turns into a vicious cycle:

  • Depression → "Nothing I do will matter."
  • Anxiety → "Everything I do will go wrong."

Scene 5: Therapy as MDP Reprogramming

Here's the hopeful side: therapy can be seen as a process of re-shaping the MDP.

  • Cognitive Behavioral Therapy (CBT):
    • Helps patients re-estimate probabilities (e.g., "Maybe the boss meeting isn't always negative").
  • Medication (SSRIs):
    • Increases baseline dopamine/serotonin, effectively raising rewards back toward normal.
  • Behavioral Activation:
    • Encourages small actions (walk, journal, meet one friend) to slowly rebuild the reward system.

Over time, the brain's MDP recalibrates:

  • Rewards feel real again.
  • Transition probabilities become realistic.
  • Value of states improves.

Scene 6: A Story of Two Paths

Consider someone with depression deciding whether to go for a walk.

  • In the depressed MDP:
    • Reward of walk = 0
    • Reward of staying in bed = 0
    • Choice: Why bother?
  • In a healthier recalibrated MDP:
    • Reward of walk = +5 (mild relief)
    • Future reward = +10 (better mood later)
    • Choice: Maybe it's worth it.

That small shift in perceived reward is the crack through which light enters the darkness.

Scene 7: Wrapping Up - Shadows Have Structure

Mental health struggles are not just about emotions. They are about the architecture of decision-making itself.

  • Depression: the reward system collapses.
  • Anxiety: the probability system exaggerates threats.

Both can be understood - and healed - by reshaping the MDP.

In Blog 5 (Finale), we'll zoom out from the individual to the collective. We'll explore how groups, societies, and even nations make choices under uncertainty. From trust to betrayal, from cooperation to competition, we'll see how multi-agent MDPs explain human social behavior.

Takeaway: Depression and anxiety are not personal failures. They are distorted decision processes. And like any MDP, with the right interventions, they can be recalibrated.

Comments

Post a Comment